release: v1.18.21

2026-08-23 22:23:18 -04:00 · 2026-08-21 14:50:55 +00:00
42 changed files with 202 additions and 668 deletions
@@ -293,7 +293,7 @@
      },
      "dependencies": {
        "@ai-sdk/alibaba": "1.0.17",
-        "@ai-sdk/amazon-bedrock": "4.0.158",
+        "@ai-sdk/amazon-bedrock": "4.0.112",
        "@ai-sdk/anthropic": "3.0.82",
        "@ai-sdk/azure": "3.0.88",
        "@ai-sdk/cerebras": "2.0.41",
@@ -568,7 +568,7 @@
        "@actions/github": "6.0.1",
        "@agentclientprotocol/sdk": "0.21.0",
        "@ai-sdk/alibaba": "1.0.17",
-        "@ai-sdk/amazon-bedrock": "4.0.158",
+        "@ai-sdk/amazon-bedrock": "4.0.112",
        "@ai-sdk/anthropic": "3.0.82",
        "@ai-sdk/azure": "3.0.88",
        "@ai-sdk/cerebras": "2.0.60",
@@ -1173,7 +1173,7 @@

    "@ai-sdk/alibaba": ["@ai-sdk/alibaba@1.0.17", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.41", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.23" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-ZbE+U5bWz2JBc5DERLowx5+TKbjGBE93LqKZAWvuEn7HOSQMraxFMZuc0ST335QZJAyfBOzh7m1mPQ+y7EaaoA=="],

-    "@ai-sdk/amazon-bedrock": ["@ai-sdk/amazon-bedrock@4.0.158", "", { "dependencies": { "@ai-sdk/anthropic": "3.0.111", "@ai-sdk/openai": "3.0.98", "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46", "@smithy/eventstream-codec": "^4.0.1", "@smithy/util-utf8": "^4.0.0", "aws4fetch": "^1.0.20" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-yZebHEszUzPLsK+Rq5sVZJkJj7EYDgY+Lz36IGf/RSkSC5LOMDbVKoQ55S8xNzJqVjUDqlWuZiQzg40HQslmCw=="],
+    "@ai-sdk/amazon-bedrock": ["@ai-sdk/amazon-bedrock@4.0.112", "", { "dependencies": { "@ai-sdk/anthropic": "3.0.81", "@ai-sdk/openai": "3.0.67", "@ai-sdk/provider": "3.0.10", "@ai-sdk/provider-utils": "4.0.27", "@smithy/eventstream-codec": "^4.0.1", "@smithy/util-utf8": "^4.0.0", "aws4fetch": "^1.0.20" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-PsSh7a6qW+3kQXPs1kD4wDwuZby0t1PIaB6j/1aMKmPFJ5LxcIcULLMF/bjITLt5o/8lc0t6TXIwG0zlhH7uZw=="],

    "@ai-sdk/anthropic": ["@ai-sdk/anthropic@3.0.82", "", { "dependencies": { "@ai-sdk/provider": "3.0.10", "@ai-sdk/provider-utils": "4.0.27" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-WKKou2wbhGGYV8PSALAPyV2YY4nfCqCPkyBzYtJtDA9yCcIFwsbtkTNgg7bqtLCVzeEsY7wwxRoCWy+EMfrw/A=="],

@@ -5641,13 +5641,13 @@

    "@ai-sdk/alibaba/@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.41", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.23" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-kNAGINk71AlOXx10Dq/PXw4t/9XjdK8uxfpVElRwtSFMdeSiLVt58p9TPx4/FJD+hxZuVhvxYj9r42osxWq79g=="],

-    "@ai-sdk/amazon-bedrock/@ai-sdk/anthropic": ["@ai-sdk/anthropic@3.0.111", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-atgBW8jZPr/KuaKX5FvDIHuXBI8VCol6kVeoD4P0657+VXR73QsLogXQVN/Zt5FHtq9WzpdIZseCJiXPqkgwwA=="],
+    "@ai-sdk/amazon-bedrock/@ai-sdk/anthropic": ["@ai-sdk/anthropic@3.0.81", "", { "dependencies": { "@ai-sdk/provider": "3.0.10", "@ai-sdk/provider-utils": "4.0.27" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-B1JDd9Ugq9R5AgIaW3674lhGCMMYJcPUxnrZh8fzbGojgg4QvHFRv6eZahGQAUsmGHbcf74G9bdSBDLWQGY2GA=="],

-    "@ai-sdk/amazon-bedrock/@ai-sdk/openai": ["@ai-sdk/openai@3.0.98", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-nAvp8pVOUJ3znJHRzZs54Y7CJQSikOW1Ty7LTdtQT+/pgtAwqhuKd8oaXbiW2xQaYBdH2i8o9AcEpcUdXIyF+g=="],
+    "@ai-sdk/amazon-bedrock/@ai-sdk/openai": ["@ai-sdk/openai@3.0.67", "", { "dependencies": { "@ai-sdk/provider": "3.0.10", "@ai-sdk/provider-utils": "4.0.27" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-oAiGC9eWG7IgtdsdS74bOCnAAHarAfTJhWN9x5INwnWPekL802AvF+0I5DvLzIF1MIRmNw4N8mPSL/GUVbX9Mw=="],

-    "@ai-sdk/amazon-bedrock/@ai-sdk/provider": ["@ai-sdk/provider@3.0.15", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-XeZW1CcDF2GMbH4wejW6xBRI2QCOgnkVYUnxoeDadB1mf85riL2bMUeDoh+6gJ/r4mjNfzUPW8OjLjvwTP0u1Q=="],
+    "@ai-sdk/amazon-bedrock/@ai-sdk/provider": ["@ai-sdk/provider@3.0.10", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-Q3BZ27qfpYqnCYGvE3vt+Qi6LGOF9R5Nmzn+9JoM1lCRsD9mYaIhfJLkSunN48nfGXJ6n+XNV0J/XVpqGQl7Dw=="],

-    "@ai-sdk/amazon-bedrock/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.46", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8", "undici": "^6.28.0" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-tEtld97plCFiYevsJuOkGkeuhQndeMWFBVrJS4AjnbD5AqrNSXRCe0p+BZ3Cju/sxDeeZ9ym3q9YUV8fASA7aQ=="],
+    "@ai-sdk/amazon-bedrock/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.27", "", { "dependencies": { "@ai-sdk/provider": "3.0.10", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-ubkAJ+xODouwtmN1tYlvTPphH1hPOBfZaEQe8U7skGvFAnIRs9PPpsq57bC2+Ky/MB4yzhd6YOsxTAx9sGpazw=="],

    "@ai-sdk/amazon-bedrock/@smithy/eventstream-codec": ["@smithy/eventstream-codec@4.2.14", "", { "dependencies": { "@aws-crypto/crc32": "5.2.0", "@smithy/types": "^4.14.1", "@smithy/util-hex-encoding": "^4.2.2", "tslib": "^2.6.2" } }, "sha512-erZq0nOIpzfeZdCyzZjdJb4nVSKLUmSkaQUVkRGQTXs30gyUGeKnrYEg+Xe1W5gE3aReS7IgsvANwVPxSzY6Pw=="],

@@ -6589,8 +6589,6 @@

    "@ai-sdk/amazon-bedrock/@ai-sdk/provider-utils/@standard-schema/spec": ["@standard-schema/spec@1.1.0", "", {}, "sha512-l2aFy5jALhniG5HgqrD6jXLi/rUWrKvqN/qJx6yoJsgKhblVd+iqqU4RCXavm/jPityDo5TCvKMnpjKnOriy0w=="],

-    "@ai-sdk/amazon-bedrock/@ai-sdk/provider-utils/undici": ["undici@6.28.0", "", {}, "sha512-LIY910g9TI13YS95lrMFrs8Rm/u/irgHeTWoKCoteeJ04CUJ92eEfj0rVn+7VKMPBpUPiUoBKfhNyLI23EE/KA=="],
-
    "@ai-sdk/anthropic/@ai-sdk/provider-utils/@standard-schema/spec": ["@standard-schema/spec@1.1.0", "", {}, "sha512-l2aFy5jALhniG5HgqrD6jXLi/rUWrKvqN/qJx6yoJsgKhblVd+iqqU4RCXavm/jPityDo5TCvKMnpjKnOriy0w=="],

    "@ai-sdk/azure/@ai-sdk/provider-utils/@standard-schema/spec": ["@standard-schema/spec@1.1.0", "", {}, "sha512-l2aFy5jALhniG5HgqrD6jXLi/rUWrKvqN/qJx6yoJsgKhblVd+iqqU4RCXavm/jPityDo5TCvKMnpjKnOriy0w=="],
@@ -1,8 +1,8 @@
 {
  "nodeModules": {
-    "x86_64-linux": "sha256-Be1I6OG6UitofhcGu2BeNzevmoQXc4Or5r/NPzwtft4=",
-    "aarch64-linux": "sha256-O+d+26CQIjZ08Rn8Qm3IytdDqIMbPdhzaGOZeUKAvIU=",
-    "aarch64-darwin": "sha256-ObS50y/oy6fM9wSGUL/wx6O0+fTWHC04mXJNd7w/2Z0=",
-    "x86_64-darwin": "sha256-eoR7ZSyH62Fq2ZaW2b2QqU2FC97rYxTMeEe+djT0nto="
+    "x86_64-linux": "sha256-yQ8EIxxYkzlEWIMY/UiIR+7lbGBuizsoehMPZzFA65Y=",
+    "aarch64-linux": "sha256-JF9VVgnl5QUZ430fqb5Qu8y0kchYJ00LO3FbYcd0lBM=",
+    "aarch64-darwin": "sha256-f3Tu6eu463NWcHgr7dupfD/zUTh26bJ6N2vVXuEyi6c=",
+    "x86_64-darwin": "sha256-miv9Sv4KdhD0UIi2O5LVS1wQOfq29IV+9S+BvIzAvgo="
  }
 }
@@ -450,7 +450,7 @@ function ModelSelectorPopoverV2View(props: {
                <For each={groups()}>
                  {(group) => (
                    <MenuV2.Group>
-                      <MenuV2.GroupLabel class="sticky top-0 z-10 gap-2 bg-v2-background-bg-layer-01 px-3">
+                      <MenuV2.GroupLabel class="gap-2 px-3">
                        <span class="min-w-0 truncate">{group.items[0].provider.name}</span>
                      </MenuV2.GroupLabel>
                      <MenuV2.RadioGroup value={props.current()}>
@@ -1,21 +0,0 @@
-import type { APIEvent } from "@solidjs/start/server"
-import { Workspace } from "@opencode-ai/console-core/workspace.js"
-import { safeEqual } from "@opencode-ai/console-core/util/crypto.js"
-import { Resource } from "@opencode-ai/console-resource"
-import z from "zod"
-
-const Body = z.object({ workspaceID: z.string().startsWith("wrk_") })
-
-export async function POST(event: APIEvent) {
-  if (!safeEqual(event.request.headers.get("authorization") ?? "", `Bearer ${Resource.SUPPORT_API_KEY.value}`)) {
-    return Response.json({ error: "Unauthorized" }, { status: 401 })
-  }
-
-  const body = Body.safeParse(await event.request.json().catch(() => undefined))
-  if (!body.success) {
-    return Response.json({ error: "Invalid request", issues: body.error.issues }, { status: 400 })
-  }
-  return Workspace.unblock(body.data)
-    .then(() => Response.json({ success: true, message: "Workspace unblocked" }))
-    .catch((error) => Response.json({ error: error instanceof Error ? error.message : String(error) }, { status: 400 }))
-}
@@ -1,5 +1,4 @@
 import type { APIEvent } from "@solidjs/start"
-import { waitUntil } from "@opencode-ai/console-resource"
 import type { DownloadPlatform } from "../types"

 const prodAssetNames: Record<string, string> = {
@@ -31,34 +30,14 @@ export async function GET({ params: { platform, channel } }: APIEvent) {
  const assetName = channel === "stable" ? prodAssetNames[platform] : betaAssetNames[platform]
  if (!assetName) return new Response(null, { status: 404 })

-  const latest = await fetch(
+  const resp = await fetch(
    `https://github.com/anomalyco/${channel === "stable" ? "opencode" : "opencode-beta"}/releases/latest/download/${assetName}`,
-    { redirect: "manual" },
  )
-  const location = latest.headers.get("location")
-  if (!location) return new Response(null, { status: 502 })

-  const key = new Request(location)
-  const cache = (caches as CacheStorage & { default: Cache }).default
-  const cached = await cache.match(key)
-  if (cached) return download(cached, platform, "HIT")
-
-  const resp = await fetch(location)
-  if (!resp.ok) return resp
-
-  const headers = new Headers(resp.headers)
-  headers.set("cache-control", "public, max-age=31536000, immutable")
-  headers.delete("set-cookie")
-  const result = new Response(resp.body, { status: resp.status, statusText: resp.statusText, headers })
-  waitUntil(cache.put(key, result.clone()))
-  return download(result, platform, "MISS")
-}
-
-function download(resp: Response, platform: string, cache: "HIT" | "MISS") {
  const downloadName = downloadNames[platform]
+
  const headers = new Headers(resp.headers)
  if (downloadName) headers.set("content-disposition", `attachment; filename="${downloadName}"`)
-  headers.set("x-opencode-cache", cache)

  return new Response(resp.body, { status: resp.status, statusText: resp.statusText, headers })
 }
@@ -28,7 +28,13 @@ import {
  GoUsageLimitError,
  BlackUsageLimitError,
 } from "./error"
-import { buildCostChunk, createStreamPartConverter, createResponseConverter, UsageInfo } from "./provider/provider"
+import {
+  buildCostChunk,
+  createBodyConverter,
+  createStreamPartConverter,
+  createResponseConverter,
+  UsageInfo,
+} from "./provider/provider"
 import { anthropicHelper } from "./provider/anthropic"
 import { googleHelper } from "./provider/google"
 import { openaiHelper } from "./provider/openai"
@@ -47,10 +53,12 @@ import { createProviderBudgetTracker } from "./providerBudgetTracker"
 import { accumulateUsage, HOT_WORKSPACES } from "./usageBatcher"
 import { Workspace } from "@opencode-ai/console-core/workspace.js"
 import { countryFromRequest, isModelCountryRestricted } from "~/lib/request-country"
-import { prepareRequestBody } from "./requestBody"

 type ZenData = Awaited<ReturnType<typeof ZenData.list>>
-type PreparedBody = Awaited<ReturnType<typeof prepareRequestBody>>
+type RetryOptions = {
+  excludeProviders: string[]
+  retryCount: number
+}
 type BillingSource = "anonymous" | "free" | "byok" | "subscription" | "lite" | "balance"

 function resolve(text: string, params?: Record<string, string | number>) {
@@ -78,6 +86,8 @@ export async function handler(
  type ProviderInfo = Awaited<ReturnType<typeof selectProvider>>
  type CostInfo = ReturnType<typeof calculateCost>

+  const MAX_FAILOVER_RETRIES = 3
+  const MAX_RETRYABLE_STATUS_RETRIES = 3
  const dict = i18n(localeFromRequest(input.request))
  const t = (key: Key, params?: Record<string, string | number>) => resolve(dict[key], params)
  const ADMIN_WORKSPACES = [
@@ -86,14 +96,12 @@ export async function handler(
    "wrk_01KKZDKDWCS1VTJF8QTX62DD50", // contributors
  ]

-  let requestBody: PreparedBody | undefined
  try {
    const url = input.request.url
-    const body = input.request.body
-    if (!body) throw new Error("Missing request body")
-    requestBody = opts.format === "google" ? undefined : await prepareRequestBody(body)
-    const model = opts.format === "google" ? opts.parseModel(url, undefined) : (requestBody?.model ?? "")
-    const googleStream = opts.format === "google" ? opts.parseIsStream(url, undefined) : undefined
+    const body = await input.request.json()
+    const model = opts.parseModel(url, body)
+    const variant = opts.parseVariant(url, body)
+    const isStream = opts.parseIsStream(url, body)
    const rawIp = input.request.headers.get("x-real-ip") ?? ""
    const ip = rawIp.includes(":") ? rawIp.split(":").slice(0, 4).join(":") : rawIp
    const rawZenApiKey = opts.parseApiKey(input.request.headers)
@@ -104,10 +112,12 @@ export async function handler(
    const projectId = input.request.headers.get("x-opencode-project") ?? ""
    const userAgent = input.request.headers.get("user-agent") ?? ""
    logger.metric({
+      is_stream: isStream,
      session: sessionId,
      request: requestId,
      client: ocClient,
      user_agent: userAgent,
+      "model.variant": variant,
      "model.tier": opts.modelList === "full" ? "zen" : "go",
    })
    const zenData = ZenData.list(opts.modelList)
@@ -165,7 +175,7 @@ export async function handler(
    )
    const providerBudget = await providerBudgetTracker?.check()

-    const providerRequest = async () => {
+    const retriableRequest = async (retry: RetryOptions = { excludeProviders: [], retryCount: 0 }) => {
      const providerInfo = selectProvider(
        model,
        zenData,
@@ -173,6 +183,7 @@ export async function handler(
        modelInfo,
        stickyId,
        trialProviders,
+        retry,
        stickyProvider,
        modelTpmLimits,
        modelTpsLimits,
@@ -188,65 +199,80 @@ export async function handler(
      })

      const startTimestamp = Date.now()
-      const reqUrl = providerInfo.modifyUrl(providerInfo.api, googleStream ?? false)
-      const specialAnthropic =
-        providerInfo.format === "anthropic" &&
-        (providerInfo.model.startsWith("arn:aws:bedrock:") ||
-          providerInfo.model.startsWith("global.anthropic.") ||
-          providerInfo.model.startsWith("databricks-claude-"))
-      if (providerInfo.format !== opts.format) throw new Error("Zen provider format must match request format")
-      if (specialAnthropic) throw new Error("Anthropic provider body modifiers are incompatible with streaming")
-      const prepared = requestBody
-
-      const reqBody = (() => {
-        if (opts.format === "google") return body
-        if (!prepared) throw new Error("Missing prepared request body")
-        return prepared.stream(providerInfo.model, providerInfo.format === "oa-compat")
-      })()
+      const reqUrl = providerInfo.modifyUrl(providerInfo.api, isStream)
+      const reqBody = JSON.stringify(
+        providerInfo.modifyBody({
+          ...createBodyConverter(opts.format, providerInfo.format)(body),
+          model: providerInfo.model,
+          ...(() => {
+            const replacer = (obj: Record<string, any>): Record<string, any> =>
+              Object.fromEntries(
+                Object.entries(obj).flatMap(([k, v]) => {
+                  if (Array.isArray(v)) return [[k, v]]
+                  if (typeof v === "object") return [[k, replacer(v)]]
+                  if (typeof v === "string") {
+                    if (v === "$workspace") return authInfo?.workspaceID ? [[k, authInfo.workspaceID]] : []
+                    if (v === "$org")
+                      return authInfo?.workspaceID ? [[k, authInfo.workspaceID.replace("wrk_", "org_")]] : []
+                    if (v === "$user") return stickyId ? [[k, stickyId]] : []
+                    if (v.startsWith("$header.")) {
+                      const headerValue = input.request.headers.get(v.slice(8))
+                      return headerValue ? [[k, headerValue]] : []
+                    }
+                  }
+                  return [[k, v]]
+                }),
+              )
+            return replacer(providerInfo.payloadModifier ?? {})
+          })(),
+        }),
+      )
      logger.debug("REQUEST URL: " + reqUrl)
-      logger.debug("REQUEST: " + (requestBody?.preview ?? "") + "...")
+      logger.debug("REQUEST: " + reqBody.substring(0, 300) + "...")
      const isNewInference =
        providerInfo.id.startsWith("console.") ||
        providerInfo.id.startsWith("console-go.") ||
        providerInfo.id.startsWith("inf.") ||
        providerInfo.id.startsWith("inf-go.")
-      const res = await fetch(reqUrl, {
-        method: "POST",
-        headers: (() => {
-          const headers = new Headers(input.request.headers)
-          providerInfo.modifyHeaders(headers, providerInfo.apiKey, stickyId)
-          Object.entries(providerInfo.headerModifier ?? {}).forEach(([k, v]) => {
-            if (v === "$ip") return headers.set(k, ip)
-            if (v === "$caller") return headers.set(k, stickyId)
-            if (v === "$session") return headers.set(k, sessionId)
-            if (v === "$model") return headers.set(k, model)
-            if (v === "$request") return headers.set(k, requestId)
-            if (v === "$project") return headers.set(k, projectId)
-            if (v === "$workspace") {
-              if (authInfo?.workspaceID) headers.set(k, authInfo.workspaceID)
-              return
-            }
-            if (v === "$org") {
-              if (authInfo?.workspaceID) headers.set(k, authInfo.workspaceID.replace("wrk_", "org_"))
-              return
-            }
-            headers.set(k, v)
-          })
-          headers.delete("host")
-          headers.delete("content-length")
-          headers.delete("x-opencode-request")
-          if (!isNewInference) headers.delete("x-opencode-session")
-          headers.delete("x-opencode-project")
-          headers.delete("x-opencode-client")
-          return headers
-        })(),
-        body: reqBody,
-        // Propagate caller disconnects to the upstream provider request so
-        // abandoned Console requests do not leave orphaned inference work open.
-        signal: input.request.signal,
-      })
-      const isStream = res.headers.get("content-type")?.toLowerCase().includes("text/event-stream") ?? false
-      logger.metric({ is_stream: isStream })
+      const res = await fetchWithRetryableStatus(
+        reqUrl,
+        {
+          method: "POST",
+          headers: (() => {
+            const headers = new Headers(input.request.headers)
+            providerInfo.modifyHeaders(headers, providerInfo.apiKey, stickyId)
+            Object.entries(providerInfo.headerModifier ?? {}).forEach(([k, v]) => {
+              if (v === "$ip") return headers.set(k, ip)
+              if (v === "$caller") return headers.set(k, stickyId)
+              if (v === "$session") return headers.set(k, sessionId)
+              if (v === "$model") return headers.set(k, model)
+              if (v === "$request") return headers.set(k, requestId)
+              if (v === "$project") return headers.set(k, projectId)
+              if (v === "$workspace") {
+                if (authInfo?.workspaceID) headers.set(k, authInfo.workspaceID)
+                return
+              }
+              if (v === "$org") {
+                if (authInfo?.workspaceID) headers.set(k, authInfo.workspaceID.replace("wrk_", "org_"))
+                return
+              }
+              headers.set(k, v)
+            })
+            headers.delete("host")
+            headers.delete("content-length")
+            headers.delete("x-opencode-request")
+            if (!isNewInference) headers.delete("x-opencode-session")
+            headers.delete("x-opencode-project")
+            headers.delete("x-opencode-client")
+            return headers
+          })(),
+          body: reqBody,
+          // Propagate caller disconnects to the upstream provider request so
+          // abandoned Console requests do not leave orphaned inference work open.
+          signal: input.request.signal,
+        },
+        { count: isNewInference ? MAX_RETRYABLE_STATUS_RETRIES : 0 },
+      )

      if (isNewInference) {
        const resEndpointId = res.headers.get("x-opencode-endpoint-id")
@@ -265,10 +291,29 @@ export async function handler(
        })
      }

-      return { providerInfo, res, startTimestamp, isStream }
+      // Try another provider => stop retrying if using fallback provider
+      if (
+        //!isNewInference &&
+        res.status !== 200 &&
+        // ie. 400 error is usually provider error like malformed request
+        res.status !== 400 &&
+        // ie. openai 404 error: Item with id 'msg_0ead8b004a3b165d0069436a6b6834819896da85b63b196a3f' not found.
+        !(modelInfo.id.startsWith("gpt-") && res.status === 404) &&
+        // ie. cannot change codex model providers mid-session
+        modelInfo.stickyProvider !== "strict" &&
+        modelInfo.fallbackProvider &&
+        providerInfo.id !== modelInfo.fallbackProvider
+      ) {
+        return retriableRequest({
+          excludeProviders: [...retry.excludeProviders, providerInfo.id],
+          retryCount: retry.retryCount + 1,
+        })
+      }
+
+      return { providerInfo, reqBody, res, startTimestamp }
    }

-    const { providerInfo, res, startTimestamp, isStream } = await providerRequest()
+    const { providerInfo, reqBody, res, startTimestamp } = await retriableRequest()

    // Store sticky provider
    if (res.status === 200) await stickyTracker?.set(providerInfo.id)
@@ -424,8 +469,6 @@ export async function handler(
      headers: resHeaders,
    })
  } catch (error: any) {
-    if (requestBody) void requestBody.cancel().catch(() => {})
-    else void input.request.body?.cancel().catch(() => {})
    // The caller disconnected before we finished. Because the outbound provider
    // request shares input.request.signal, an aborted caller surfaces here as an
    // AbortError. There is no client left to receive a body, so skip the error
@@ -550,6 +593,7 @@ export async function handler(
    modelInfo: ModelInfo,
    stickyId: string,
    trialProviders: string[] | undefined,
+    retry: RetryOptions,
    stickyProviderId: string | undefined,
    modelTpmLimits: Record<string, number> | undefined,
    modelTpsLimits: Record<string, { qualify: number; unqualify: number }> | undefined,
@@ -576,11 +620,14 @@ export async function handler(
        }))
      }

+      // Use fallback provider if max retries reached
      const fallbackProvider = allProviders.find((provider) => provider.id === modelInfo.fallbackProvider)
+      if (retry.retryCount === MAX_FAILOVER_RETRIES) return fallbackProvider

      let topPriority = Infinity
      const providers = allProviders
        .filter((provider) => provider.weight !== 0)
+        .filter((provider) => !retry.excludeProviders.includes(provider.id))
        .filter((provider) => {
          if (provider.budgetPriority === undefined) return true
          if (!providerBudget) return true
@@ -988,6 +1035,15 @@ export async function handler(
    providerInfo.apiKey = authInfo.provider.credentials
  }

+  async function fetchWithRetryableStatus(url: string, options: RequestInit, retry = { count: 0 }) {
+    const res = await fetch(url, options)
+    if ([429, 529].includes(res.status) && retry.count < MAX_RETRYABLE_STATUS_RETRIES) {
+      await new Promise((resolve) => setTimeout(resolve, Math.pow(2, retry.count) * 500))
+      return fetchWithRetryableStatus(url, options, { count: retry.count + 1 })
+    }
+    return res
+  }
+
  function calculateCost(modelInfo: ModelInfo, usageInfo: UsageInfo) {
    const { inputTokens, outputTokens, reasoningTokens, cacheReadTokens, cacheWrite5mTokens, cacheWrite1hTokens } =
      usageInfo
@@ -1,191 +0,0 @@
-const TAIL_LIMIT = 4 * 1024
-const encoder = new TextEncoder()
-
-export async function prepareRequestBody(body: ReadableStream<Uint8Array>) {
-  const reader = body.getReader()
-  const chunks: Uint8Array[] = []
-  const decoder = new TextDecoder()
-  let text = ""
-  let done = false
-  let searchFrom = 0
-  let bom = 0
-  let match: RegExpExecArray | null = null
-  const pattern = /("model"\s*:\s*")([^"]+)"/g
-
-  while (!done && !match) {
-    const next = await reader.read()
-    done = next.done
-    if (!next.value) continue
-    if (!chunks.length && next.value[0] === 0xef && next.value[1] === 0xbb && next.value[2] === 0xbf) bom = 3
-    chunks.push(next.value)
-    text += decoder.decode(next.value, { stream: true })
-    pattern.lastIndex = searchFrom
-    match = pattern.exec(text)
-    searchFrom = Math.max(0, text.length - 256)
-  }
-  if (done) {
-    text += decoder.decode()
-    if (!match) {
-      pattern.lastIndex = searchFrom
-      match = pattern.exec(text)
-    }
-  }
-
-  const found = (() => {
-    if (!match) return
-    const start = bom + utf8Length(text, match.index + match[1].length)
-    return { model: match[2], start, end: start + utf8Length(match[2], match[2].length) }
-  })()
-  const preview = text.substring(0, 300)
-  text = ""
-  match = null
-  let used = false
-
-  return {
-    model: found?.model ?? "",
-    preview,
-    cancel: () => reader.cancel(),
-    stream(providerModel: string, includeUsage: boolean) {
-      if (used) throw new Error("Request body stream already consumed")
-      if (!found) throw new Error("Missing model field")
-      used = true
-
-      const initial = replace(chunks, found.start, found.end, providerModel)
-      chunks.length = 0
-      const output = passthrough(initial, reader, done)
-      if (!includeUsage) return output
-      return appendUsage(output)
-    },
-  }
-}
-
-function utf8Length(value: string, end: number) {
-  let length = 0
-  for (let i = 0; i < end; i++) {
-    const code = value.charCodeAt(i)
-    if (code <= 0x7f) length++
-    else if (code <= 0x7ff) length += 2
-    else if (code >= 0xd800 && code <= 0xdbff && i + 1 < end && value.charCodeAt(i + 1) >= 0xdc00) {
-      length += 4
-      i++
-    } else length += 3
-  }
-  return length
-}
-
-function replace(chunks: Uint8Array[], start: number, end: number, value: string) {
-  let offset = 0
-  let inserted = false
-  return chunks.flatMap((chunk) => {
-    const chunkStart = offset
-    const chunkEnd = offset + chunk.length
-    offset = chunkEnd
-    if (chunkEnd <= start || chunkStart >= end) return [chunk]
-
-    const parts = [chunk.subarray(0, Math.max(0, start - chunkStart))]
-    if (!inserted) {
-      parts.push(encoder.encode(value))
-      inserted = true
-    }
-    parts.push(chunk.subarray(Math.min(chunk.length, end - chunkStart)))
-    return parts.filter((part) => part.length)
-  })
-}
-
-function passthrough(
-  initial: Array<Uint8Array | undefined>,
-  reader: ReadableStreamDefaultReader<Uint8Array>,
-  sourceDone: boolean,
-) {
-  let done = sourceDone
-  let index = 0
-  return new ReadableStream<Uint8Array>({
-    async pull(controller) {
-      const chunk = initial[index]
-      if (chunk) {
-        initial[index++] = undefined
-        controller.enqueue(chunk)
-        return
-      }
-      initial.length = 0
-      if (done) {
-        controller.close()
-        return
-      }
-      const next = await reader.read()
-      done = next.done
-      if (next.value) controller.enqueue(next.value)
-      if (done) controller.close()
-    },
-    cancel(reason) {
-      initial.length = 0
-      return reader.cancel(reason)
-    },
-  })
-}
-
-function appendUsage(body: ReadableStream<Uint8Array>) {
-  const reader = body.getReader()
-  const decoder = new TextDecoder()
-  let tail = new Uint8Array()
-  let streamText = ""
-  let isStream = false
-  const inspect = (chunk?: Uint8Array) => {
-    streamText += chunk ? decoder.decode(chunk, { stream: true }) : decoder.decode()
-    for (const match of streamText.matchAll(/"stream"\s*:\s*(true|false)/g)) isStream = match[1] === "true"
-    streamText = streamText.slice(-64)
-  }
-  return new ReadableStream<Uint8Array>({
-    async pull(controller) {
-      while (true) {
-        const next = await reader.read()
-        if (next.done) {
-          inspect()
-          if (!isStream) {
-            if (tail.length) controller.enqueue(tail)
-            controller.close()
-            return
-          }
-          const close = tail.lastIndexOf(125)
-          if (close < 0) {
-            controller.error(new Error("Invalid JSON request body"))
-            return
-          }
-          if (close) controller.enqueue(tail.subarray(0, close))
-          controller.enqueue(encoder.encode(',"stream_options":{"include_usage":true}}'))
-          if (close + 1 < tail.length) controller.enqueue(tail.subarray(close + 1))
-          controller.close()
-          return
-        }
-
-        const chunk = next.value
-        inspect(chunk)
-        if (tail.length + chunk.length <= TAIL_LIMIT) {
-          const combined = new Uint8Array(tail.length + chunk.length)
-          combined.set(tail)
-          combined.set(chunk, tail.length)
-          tail = combined
-          continue
-        }
-
-        const emit = tail.length + chunk.length - TAIL_LIMIT
-        if (emit <= tail.length) {
-          controller.enqueue(tail.subarray(0, emit))
-          const combined = new Uint8Array(TAIL_LIMIT)
-          combined.set(tail.subarray(emit))
-          combined.set(chunk, tail.length - emit)
-          tail = combined
-          return
-        }
-
-        if (tail.length) controller.enqueue(tail)
-        controller.enqueue(chunk.subarray(0, emit - tail.length))
-        tail = chunk.slice(emit - tail.length)
-        return
-      }
-    },
-    cancel(reason) {
-      return reader.cancel(reason)
-    },
-  })
-}
@@ -1,122 +0,0 @@
-import { describe, expect, test } from "bun:test"
-import { prepareRequestBody } from "../src/routes/zen/util/requestBody"
-
-describe("Zen request body streaming", () => {
-  test("patches the leading model without buffering the remaining body", async () => {
-    let reads = 0
-    const body = new ReadableStream<Uint8Array>(
-      {
-        pull(controller) {
-          const chunks = [
-            '{"model":"client-model","stream":true,"messages":[',
-            JSON.stringify({ role: "user", content: "large payload" }),
-            "]}",
-          ]
-          const chunk = chunks[reads++]
-          if (chunk) controller.enqueue(new TextEncoder().encode(chunk))
-          else controller.close()
-        },
-      },
-      { highWaterMark: 0 },
-    )
-
-    const request = await prepareRequestBody(body)
-    expect(request.model).toBe("client-model")
-    expect(reads).toBe(1)
-
-    const output = await new Response(request.stream("provider-model", false)).text()
-    expect(JSON.parse(output)).toEqual({
-      model: "provider-model",
-      stream: true,
-      messages: [{ role: "user", content: "large payload" }],
-    })
-  })
-
-  test("appends stream usage options at the end of the request", async () => {
-    const body = new Blob(['{"model":"client-model","stream":true,"messages":[]}   ']).stream()
-    const request = await prepareRequestBody(body)
-    const output = await new Response(request.stream("provider-model", true)).text()
-
-    expect(JSON.parse(output)).toEqual({
-      model: "provider-model",
-      stream: true,
-      messages: [],
-      stream_options: { include_usage: true },
-    })
-    expect(output.endsWith("   ")).toBe(true)
-  })
-
-  test("detects streaming after a large message while forwarding", async () => {
-    const content = "x".repeat(128 * 1024)
-    let reads = 0
-    const chunks = [
-      '{"model":"client-model","messages":[',
-      JSON.stringify({ role: "user", content }),
-      '],"stream":true}',
-    ]
-    const body = new ReadableStream<Uint8Array>(
-      {
-        pull(controller) {
-          const chunk = chunks[reads++]
-          if (chunk) controller.enqueue(new TextEncoder().encode(chunk))
-          else controller.close()
-        },
-      },
-      { highWaterMark: 0 },
-    )
-    const request = await prepareRequestBody(body)
-    expect(reads).toBe(1)
-    const output = await new Response(request.stream("provider-model", true)).text()
-
-    expect(JSON.parse(output)).toEqual({
-      model: "provider-model",
-      messages: [{ role: "user", content }],
-      stream: true,
-      stream_options: { include_usage: true },
-    })
-  })
-
-  test("buffers through a late model field and then streams the rest", async () => {
-    const content = "こんにちは".repeat(32 * 1024)
-    let reads = 0
-    const chunks = [
-      '{"messages":[',
-      JSON.stringify({ role: "user", content }),
-      '],"model":"client-model","stream":true,"extra":"after-model"}',
-    ]
-    const body = new ReadableStream<Uint8Array>(
-      {
-        pull(controller) {
-          const chunk = chunks[reads++]
-          if (chunk) controller.enqueue(new TextEncoder().encode(chunk))
-          else controller.close()
-        },
-      },
-      { highWaterMark: 0 },
-    )
-    const request = await prepareRequestBody(body)
-
-    expect(request.model).toBe("client-model")
-    expect(reads).toBe(3)
-    expect(JSON.parse(await new Response(request.stream("provider-model", true)).text())).toEqual({
-      messages: [{ role: "user", content }],
-      model: "provider-model",
-      stream: true,
-      extra: "after-model",
-      stream_options: { include_usage: true },
-    })
-  })
-
-  test("preserves a UTF-8 BOM while patching the model", async () => {
-    const body = new Blob(['\uFEFF{"messages":[],"model":"client-model","stream":false}']).stream()
-    const request = await prepareRequestBody(body)
-    const output = new Uint8Array(await new Response(request.stream("provider-model", false)).arrayBuffer())
-
-    expect([...output.subarray(0, 3)]).toEqual([0xef, 0xbb, 0xbf])
-    expect(JSON.parse(new TextDecoder().decode(output))).toEqual({
-      messages: [],
-      model: "provider-model",
-      stream: false,
-    })
-  })
-})
@@ -97,18 +97,6 @@ export namespace Workspace {
    },
  )

-  export const unblock = fn(
-    z.object({
-      workspaceID: Identifier.schema("workspace"),
-    }),
-    async (input) => {
-      const result = await Database.use((tx) =>
-        tx.update(WorkspaceTable).set({ is_blocked: false }).where(eq(WorkspaceTable.id, input.workspaceID)),
-      )
-      if (result.rowsAffected === 0) throw new Error("Workspace not found")
-    },
-  )
-
  export const remove = fn(z.void(), async () => {
    await Database.use((tx) =>
      tx
@@ -62,7 +62,7 @@
  },
  "dependencies": {
    "@ai-sdk/alibaba": "1.0.17",
-    "@ai-sdk/amazon-bedrock": "4.0.158",
+    "@ai-sdk/amazon-bedrock": "4.0.112",
    "@ai-sdk/anthropic": "3.0.82",
    "@ai-sdk/azure": "3.0.88",
    "@ai-sdk/cerebras": "2.0.41",
@@ -45,18 +45,9 @@ function oauth(http: HttpClient.HttpClient) {
    authorize: () =>
      Effect.gen(function* () {
        const device = yield* post(http, `${defaultServer}/auth/device/code`, { client_id: clientID }, Device)
-        const verification = yield* Effect.try({
-          try: () => {
-            const url = new URL(device.verification_uri_complete, `${defaultServer}/`)
-            if (url.protocol !== "http:" && url.protocol !== "https:") throw new Error("expected HTTP(S)")
-            return url
-          },
-          catch: (cause) =>
-            new Error(`Invalid device verification URL: ${cause instanceof Error ? cause.message : String(cause)}`),
-        })
        return {
          mode: "auto" as const,
-          url: verification.href,
+          url: `${defaultServer}${device.verification_uri_complete}`,
          instructions: `Enter code: ${device.user_code}`,
          callback: poll(http, defaultServer, device.device_code, Duration.seconds(device.interval)),
        }
@@ -1,6 +1,5 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
-import { HttpClient, HttpClientResponse } from "effect/unstable/http"
 import { Catalog } from "@opencode-ai/core/catalog"
 import { Credential } from "@opencode-ai/core/credential"
 import { EventV2 } from "@opencode-ai/core/event"
@@ -15,16 +14,14 @@ import { PluginTestLayer } from "./fixture"

 const it = testEffect(PluginTestLayer)

-const addPlugin = Effect.fn(function* (http?: HttpClient.HttpClient) {
+const addPlugin = Effect.fn(function* () {
  const plugin = yield* PluginV2.Service
  const host = yield* PluginHost.make(plugin)
  const events = yield* EventV2.Service
  const integration = yield* Integration.Service
-  const client = yield* HttpClient.HttpClient
  yield* OpencodePlugin.effect(host).pipe(
    Effect.provideService(EventV2.Service, events),
    Effect.provideService(Integration.Service, integration),
-    Effect.provideService(HttpClient.HttpClient, http ?? client),
  )
 })

@@ -85,63 +82,6 @@ describe("OpencodePlugin", () => {
    }),
  )

-  it.effect("resolves origin-rooted device verification URLs", () =>
-    Effect.gen(function* () {
-      const http = HttpClient.make((request) =>
-        Effect.succeed(
-          HttpClientResponse.fromWeb(
-            request,
-            Response.json({
-              device_code: "device",
-              user_code: "user",
-              verification_uri_complete: "/console/device?user_code=user&client_id=opencode-cli",
-              expires_in: 60,
-              interval: 60,
-            }),
-          ),
-        ),
-      )
-      yield* addPlugin(http)
-      const integration = yield* Integration.Service
-      const attempt = yield* integration.connection.oauth({
-        integrationID: Integration.ID.make("opencode"),
-        methodID: Integration.MethodID.make("device"),
-        inputs: {},
-      })
-      expect(attempt.url).toBe("https://opencode.ai/console/device?user_code=user&client_id=opencode-cli")
-    }),
-  )
-
-  it.effect("rejects malformed device verification URLs", () =>
-    Effect.gen(function* () {
-      const http = HttpClient.make((request) =>
-        Effect.succeed(
-          HttpClientResponse.fromWeb(
-            request,
-            Response.json({
-              device_code: "device",
-              user_code: "user",
-              verification_uri_complete: "http://[::1",
-              expires_in: 60,
-              interval: 60,
-            }),
-          ),
-        ),
-      )
-      yield* addPlugin(http)
-      const integration = yield* Integration.Service
-      const error = yield* integration.connection
-        .oauth({
-          integrationID: Integration.ID.make("opencode"),
-          methodID: Integration.MethodID.make("device"),
-          inputs: {},
-        })
-        .pipe(Effect.flip)
-      expect(error).toBeInstanceOf(Integration.AuthorizationError)
-      expect(String(error.cause)).toContain("Invalid device verification URL")
-    }),
-  )
-
  it.live("loads providers and models from the connected OpenCode server", () =>
    Effect.acquireUseRelease(
      Effect.sync(() => {
@@ -56,7 +56,7 @@
    "@actions/github": "6.0.1",
    "@agentclientprotocol/sdk": "0.21.0",
    "@ai-sdk/alibaba": "1.0.17",
-    "@ai-sdk/amazon-bedrock": "4.0.158",
+    "@ai-sdk/amazon-bedrock": "4.0.112",
    "@ai-sdk/anthropic": "3.0.82",
    "@ai-sdk/azure": "3.0.88",
    "@ai-sdk/cerebras": "2.0.60",
@@ -396,18 +396,10 @@ const layer: Layer.Layer<Service, never, AccountRepo.Service | HttpClient.HttpCl
      const parsed = yield* HttpClientResponse.schemaBodyJson(DeviceAuth)(response).pipe(
        mapAccountServiceError("Failed to decode response"),
      )
-      const verification = yield* Effect.try({
-        try: () => {
-          const url = new URL(parsed.verification_uri_complete, `${normalizedServer}/`)
-          if (url.protocol !== "http:" && url.protocol !== "https:") throw new Error("expected HTTP(S)")
-          return url.href
-        },
-        catch: (cause) => new AccountServiceError({ message: "Invalid device verification URL", cause }),
-      })
      return new Login({
        code: parsed.device_code,
        user: parsed.user_code,
-        url: verification,
+        url: `${normalizedServer}${parsed.verification_uri_complete}`,
        server: normalizedServer,
        expiry: parsed.expires_in,
        interval: parsed.interval,
@@ -1303,13 +1303,13 @@ export function options(input: {
      }
    }

-    // Generic OpenAI-compatible APIs do not necessarily support OpenAI's verbosity parameter.
-    // Only enable the default for integrations known to implement it.
+    // Only set textVerbosity for non-chat gpt-5.x models
+    // Chat models (e.g. gpt-5.2-chat-latest) only support "medium" verbosity
    if (
      input.model.api.id.includes("gpt-5.") &&
      !input.model.api.id.includes("codex") &&
      !input.model.api.id.includes("-chat") &&
-      (input.model.api.npm === "@ai-sdk/openai" || input.model.api.npm === "@ai-sdk/amazon-bedrock/mantle")
+      input.model.providerID !== "azure"
    ) {
      result["textVerbosity"] = "low"
    }
@@ -10,7 +10,6 @@ import { Account } from "../../src/account/account"
 import {
  AccessToken,
  AccountID,
-  AccountServiceError,
  AccountTransportError,
  DeviceCode,
  Login,
@@ -72,18 +71,18 @@ const deviceTokenClient = (body: unknown, status = 400) =>
 const poll = (body: unknown, status = 400) =>
  Account.Service.use((s) => s.poll(login())).pipe(Effect.provide(live(deviceTokenClient(body, status))))

-it.live("login resolves origin-rooted verification URLs from servers with base paths", () =>
+it.live("login normalizes trailing slashes in the provided server URL", () =>
  Effect.gen(function* () {
    const seen: Array<string> = []
    const client = HttpClient.make((req) =>
      Effect.gen(function* () {
        seen.push(`${req.method} ${req.url}`)

-        if (req.url === "https://one.example.com/console/auth/device/code") {
+        if (req.url === "https://one.example.com/auth/device/code") {
          return json(req, {
            device_code: "device-code",
            user_code: "user-code",
-            verification_uri_complete: "/console/device?user_code=user-code",
+            verification_uri_complete: "/device?user_code=user-code",
            expires_in: 600,
            interval: 5,
          })
@@ -93,31 +92,11 @@ it.live("login resolves origin-rooted verification URLs from servers with base p
      }),
    )

-    const result = yield* Account.use.login("https://one.example.com/console/").pipe(Effect.provide(live(client)))
+    const result = yield* Account.use.login("https://one.example.com/").pipe(Effect.provide(live(client)))

-    expect(seen).toEqual(["POST https://one.example.com/console/auth/device/code"])
-    expect(result.server).toBe("https://one.example.com/console")
-    expect(result.url).toBe("https://one.example.com/console/device?user_code=user-code")
-  }),
-)
-
-it.live("login rejects malformed device verification URLs", () =>
-  Effect.gen(function* () {
-    const client = HttpClient.make((req) =>
-      Effect.succeed(
-        json(req, {
-          device_code: "device-code",
-          user_code: "user-code",
-          verification_uri_complete: "http://[::1",
-          expires_in: 600,
-          interval: 5,
-        }),
-      ),
-    )
-
-    const error = yield* Effect.flip(Account.use.login("https://one.example.com").pipe(Effect.provide(live(client))))
-    expect(error).toBeInstanceOf(AccountServiceError)
-    if (error instanceof AccountServiceError) expect(error.message).toBe("Invalid device verification URL")
+    expect(seen).toEqual(["POST https://one.example.com/auth/device/code"])
+    expect(result.server).toBe("https://one.example.com")
+    expect(result.url).toBe("https://one.example.com/device?user_code=user-code")
  }),
 )

@@ -82,10 +82,10 @@ describe("opencode run (non-interactive subprocess)", () => {
  )

  // The test provider's SSE error item is interpreted by the SDK as an unknown
-  // finish, not a fatal provider/session error. Unknown finishes should continue
-  // the prompt loop so a subsequent response can complete the run.
+  // finish, not a fatal provider/session error. Lock that distinction in so it
+  // is not accidentally used as the failure compatibility oracle.
  cliIt.concurrent(
-    "unknown stream finish preserves partial output and continues",
+    "unknown stream finish preserves partial output and exits 0",
    ({ llm, opencode }) =>
      Effect.gen(function* () {
        yield* llm.push(
@@ -95,10 +95,9 @@ describe("opencode run (non-interactive subprocess)", () => {
          }),
        )
        yield* llm.fail("upstream provider exploded mid-stream")
-        yield* llm.text("recovered")
        const result = yield* opencode.run("trigger midstream error", { timeoutMs: 30_000 })
        expect(result.exitCode).toBe(0)
-        expect(result.stdout).toBe("partial response\nrecovered\n")
+        expect(result.stdout).toBe("partial response\n")
        expect(result.stderr).not.toContain("upstream provider exploded mid-stream")
      }),
    60_000,
@@ -214,7 +213,7 @@ describe("opencode run (non-interactive subprocess)", () => {
  )

  cliIt.concurrent(
-    "--format json records an unknown stream finish and continuation",
+    "--format json records partial output for an unknown stream finish",
    ({ llm, opencode }) =>
      Effect.gen(function* () {
        yield* llm.push(
@@ -224,7 +223,6 @@ describe("opencode run (non-interactive subprocess)", () => {
          }),
        )
        yield* llm.fail("provider failed")
-        yield* llm.text("recovered")
        const result = yield* opencode.run("fail after output", { format: "json" })

        const events = opencode.parseJsonEvents(result.stdout)
@@ -236,14 +234,9 @@ describe("opencode run (non-interactive subprocess)", () => {
          "step_finish",
          "step_start",
          "step_finish",
-          "step_start",
-          "text",
-          "step_finish",
        ])
        expect(events[1]?.part).toEqual(expect.objectContaining({ type: "text", text: "partial json" }))
-        expect(events[5]?.part).toEqual(expect.objectContaining({ type: "step-finish", reason: "unknown" }))
-        expect(events[7]?.part).toEqual(expect.objectContaining({ type: "text", text: "recovered" }))
-        expect(events.at(-1)?.part).toEqual(expect.objectContaining({ type: "step-finish", reason: "stop" }))
+        expect(events.at(-1)?.part).toEqual(expect.objectContaining({ type: "step-finish", reason: "unknown" }))
      }),
    60_000,
  )
@@ -523,7 +523,6 @@ describe("ProviderTransform.options - gpt-5 textVerbosity", () => {
    expect(result.reasoningEffort).toBe("medium")
    expect(result.reasoningSummary).toBeUndefined()
    expect(result.include).toBeUndefined()
-    expect(result.textVerbosity).toBeUndefined()
  })

  test("azure chat completions omit Responses-only reasoning options after variants merge", async () => {
@@ -812,8 +812,7 @@ function ModelTrendSection(props: {
                  <div data-slot="tooltip-divider" />
                  <p>
                    <span data-slot="tooltip-label">
-                      <i data-kind={props.lineTone === "active" ? "users" : "tokens"} />
-                      <span data-slot="tooltip-name">{props.rowLabel}</span>
+                      <i data-kind={props.lineTone === "active" ? "users" : "tokens"} /> {props.rowLabel}
                    </span>
                    <b>{props.formatValue(props.value(active.point))}</b>
                  </p>
@@ -473,15 +473,13 @@ function LabUsageSection(props: { lab: ModelCatalogLab; data: StatsLabData | nul
                    <div data-slot="tooltip-divider" />
                    <p>
                      <span data-slot="tooltip-label">
-                        <i data-kind="tokens" />
-                        <span data-slot="tooltip-name">{i18n.t("lab.dailyTokens")}</span>
+                        <i data-kind="tokens" /> {i18n.t("lab.dailyTokens")}
                      </span>
                      <b>{formatTokens(active.point.tokens)}</b>
                    </p>
                    <p>
                      <span data-slot="tooltip-label">
-                        <i data-kind="users" />
-                        <span data-slot="tooltip-name">{i18n.t("model.uniqueUsers")}</span>
+                        <i data-kind="users" /> {i18n.t("model.uniqueUsers")}
                      </span>
                      <b>{formatUsers(active.point.users)}</b>
                    </p>
@@ -1590,13 +1590,6 @@ body {
  white-space: nowrap;
 }

-[data-page="stats"] [data-component="chart-tooltip"] [data-slot="tooltip-name"] {
-  min-width: 0;
-  overflow: hidden;
-  text-overflow: ellipsis;
-  white-space: nowrap;
-}
-
 [data-page="stats"] [data-component="chart-tooltip"] [data-slot="tooltip-divider"] {
  height: 1px;
  margin: 4px -16px 2px;
@@ -1684,7 +1677,7 @@ body {

 [data-page="stats"] :is([data-section="top-models"], [data-section="unique-users"]) [data-component="chart-tooltip"] p {
  grid-template-columns: minmax(0, 1fr) auto;
-  gap: 1ch;
+  gap: 4px;
  height: 16px;
  margin: 4px 0 0;
  padding: 0 8px;
@@ -4730,7 +4723,7 @@ body {

 [data-page="stats"] [data-component="model-usage-chart"] [data-component="chart-tooltip"] p {
  grid-template-columns: minmax(0, 1fr) auto;
-  gap: 1ch;
+  gap: 4px;
  height: 20px;
  margin: 8px 0;
  padding: 0 8px;
@@ -568,8 +568,8 @@ function TopModelsChart(props: {
                              style={{
                                background: getRankColor(item.segment.model, item.index, segmentOrder(), usageColors),
                              }}
-                            />
-                            <span data-slot="tooltip-name">{item.segment.model}</span>
+                            />{" "}
+                            {item.segment.model}
                          </span>
                          <b>{formatUsageChartValue(item.segment.value, metric())}</b>
                        </p>
@@ -4,7 +4,6 @@ import { modelAuthor, normalizeInferenceModel, statModel, statProvider } from ".

 describe("inference stat normalization", () => {
  test("normalizes model suffixes used by router/provider variants", () => {
-    expect(normalizeInferenceModel("GPT-5-Free")).toBe("gpt-5")
    expect(normalizeInferenceModel("deepseek-v4-flash-free")).toBe("deepseek-v4-flash")
    expect(normalizeInferenceModel("deepseek-v4-flash:global")).toBe("deepseek-v4-flash")
    expect(normalizeInferenceModel("mimo-v2.5-free")).toBe("mimo-v2.5")
@@ -23,7 +23,7 @@ export const RETIRED_STAT_MODELS = ["big-pickle", ...Object.keys(MODEL_NAME_ALIA
 export const RETIRED_STAT_PROVIDERS = ["opencode"]

 export function normalizeInferenceModel(value: string | undefined) {
-  return (value || "unknown").toLowerCase().replace(/(-free|:free|:global)+$/, "") || "unknown"
+  return (value || "unknown").replace(/(-free|:free|:global)+$/, "") || "unknown"
 }

 export function modelAuthor(value: string | undefined) {
@@ -188,8 +188,8 @@ https://opencode.ai/zen/v1/models
 | Grok 4.5 (> 200K tokens)          | $4.00   | $12.00  | $0.60           | -               |
 | Grok Build 0.1                    | $1.00   | $2.00   | $0.20           | -               |
 | Muse Spark 1.2                    | $1.25   | $4.25   | $0.15           | -               |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00   | $10.00  | $0.20           | $2.50           |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00   | $15.00  | $0.40           | $5.00           |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00   | $30.00  | $0.50           | $6.25           |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00  | $45.00  | $1.00           | $12.50          |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00   | $12.00  | $0.20           | $2.50           |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00   | $18.00  | $0.40           | $5.00           |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20   | $1.20   | $0.02           | $0.25           |
@@ -214,8 +214,6 @@ https://opencode.ai/zen/v1/models
 | GPT 5 Codex                       | $1.07   | $8.50   | $0.107          | -               |
 | GPT 5 Nano                        | $0.05   | $0.40   | $0.005          | -               |

-**GPT 5.6 Sol:** تشمل الأسعار المعروضة خصمًا بنسبة 50% حتى 18 سبتمبر 2026.
-
 **DeepSeek V4 Flash / Pro:** ساعات Peak هي 01:00-04:00 و06:00-10:00 UTC؛ وجميع الساعات الأخرى Off-Peak. [اعرف المزيد](https://api-docs.deepseek.com/quick_start/pricing/).

 قد تلاحظ [نماذج منخفضة التكلفة](/docs/config/#models)، مثل Haiku أو Nano أو Flash، في سجل الاستخدام. يستخدم OpenCode هذه النماذج لإنشاء عناوين الجلسات.
@@ -195,8 +195,8 @@ Podržavamo pay-as-you-go model. Ispod su cijene **po 1M tokena**.
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -221,8 +221,6 @@ Podržavamo pay-as-you-go model. Ispod su cijene **po 1M tokena**.
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** Prikazane cijene uključuju popust od 50% do 18. septembra 2026.
-
 **DeepSeek V4 Flash / Pro:** Peak sati su 01:00-04:00 i 06:00-10:00 UTC; svi ostali sati su Off-Peak. [Saznajte više](https://api-docs.deepseek.com/quick_start/pricing/).

 U historiji korištenja možete primijetiti [jeftinije modele](/docs/config/#models), kao što su Haiku, Nano ili Flash. OpenCode koristi ove modele za generisanje naslova sesija.
@@ -195,8 +195,8 @@ Vi understøtter en pay-as-you-go-model. Nedenfor er priserne **pr. 1M tokens**.
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -221,8 +221,6 @@ Vi understøtter en pay-as-you-go-model. Nedenfor er priserne **pr. 1M tokens**.
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** De viste priser inkluderer 50 % rabat til og med den 18. september 2026.
-
 **DeepSeek V4 Flash / Pro:** Peak-tiderne er 01:00-04:00 og 06:00-10:00 UTC; alle andre tider er Off-Peak. [Læs mere](https://api-docs.deepseek.com/quick_start/pricing/).

 Du vil måske bemærke [lavprismodeller](/docs/config/#models), såsom Haiku, Nano eller Flash, i din brugshistorik. OpenCode bruger disse modeller til at generere sessionstitler.
@@ -184,8 +184,8 @@ Wir unterstützen ein Pay-as-you-go-Modell. Unten findest du die Preise **pro 1M
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -210,8 +210,6 @@ Wir unterstützen ein Pay-as-you-go-Modell. Unten findest du die Preise **pro 1M
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** Die angezeigten Preise enthalten bis zum 18. September 2026 einen Rabatt von 50 %.
-
 **DeepSeek V4 Flash / Pro:** Die Peak-Zeiten sind 01:00-04:00 und 06:00-10:00 UTC; alle anderen Zeiten sind Off-Peak. [Mehr erfahren](https://api-docs.deepseek.com/quick_start/pricing/).

 Möglicherweise siehst du [kostengünstige Modelle](/docs/config/#models) wie Haiku, Nano oder Flash in deinem Nutzungsverlauf. OpenCode verwendet diese Modelle, um Session-Titel zu generieren.
@@ -195,8 +195,8 @@ Admitimos un modelo de pago por uso. A continuación se muestran los precios **p
 | Grok 4.5 (> 200K tokens)          | $4.00   | $12.00  | $0.60            | -                  |
 | Grok Build 0.1                    | $1.00   | $2.00   | $0.20            | -                  |
 | Muse Spark 1.2                    | $1.25   | $4.25   | $0.15            | -                  |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00   | $10.00  | $0.20            | $2.50              |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00   | $15.00  | $0.40            | $5.00              |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00   | $30.00  | $0.50            | $6.25              |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00  | $45.00  | $1.00            | $12.50             |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00   | $12.00  | $0.20            | $2.50              |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00   | $18.00  | $0.40            | $5.00              |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20   | $1.20   | $0.02            | $0.25              |
@@ -221,8 +221,6 @@ Admitimos un modelo de pago por uso. A continuación se muestran los precios **p
 | GPT 5 Codex                       | $1.07   | $8.50   | $0.107           | -                  |
 | GPT 5 Nano                        | $0.05   | $0.40   | $0.005           | -                  |

-**GPT 5.6 Sol:** Los precios mostrados incluyen un 50 % de descuento hasta el 18 de septiembre de 2026.
-
 **DeepSeek V4 Flash / Pro:** Las horas Peak son 01:00-04:00 y 06:00-10:00 UTC; todas las demás horas son Off-Peak. [Más información](https://api-docs.deepseek.com/quick_start/pricing/).

 Puede que notes [modelos de bajo costo](/docs/config/#models), como Haiku, Nano o Flash, en tu historial de uso. OpenCode usa estos modelos para generar títulos de sesiones.
@@ -184,8 +184,8 @@ Nous prenons en charge un modèle de paiement à l'utilisation. Vous trouverez c
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -210,8 +210,6 @@ Nous prenons en charge un modèle de paiement à l'utilisation. Vous trouverez c
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** Les prix affichés incluent une réduction de 50 % jusqu’au 18 septembre 2026.
-
 **DeepSeek V4 Flash / Pro:** Les heures Peak sont 01:00-04:00 et 06:00-10:00 UTC ; toutes les autres heures sont Off-Peak. [En savoir plus](https://api-docs.deepseek.com/quick_start/pricing/).

 Vous remarquerez peut-être des [modèles à faible coût](/docs/config/#models), tels que Haiku, Nano ou Flash, dans votre historique d'utilisation. OpenCode utilise ces modèles pour générer les titres des sessions.
@@ -195,8 +195,8 @@ Supportiamo un modello pay-as-you-go. Qui sotto trovi i prezzi **per 1M token**.
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -221,8 +221,6 @@ Supportiamo un modello pay-as-you-go. Qui sotto trovi i prezzi **per 1M token**.
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** I prezzi mostrati includono uno sconto del 50% fino al 18 septembre 2026.
-
 **DeepSeek V4 Flash / Pro:** Gli orari Peak sono 01:00-04:00 e 06:00-10:00 UTC; tutti gli altri orari sono Off-Peak. [Scopri di più](https://api-docs.deepseek.com/quick_start/pricing/).

 Potresti notare [modelli a basso costo](/docs/config/#models), come Haiku, Nano o Flash, nella cronologia di utilizzo. OpenCode usa questi modelli per generare i titoli delle sessioni.
@@ -184,8 +184,8 @@ https://opencode.ai/zen/v1/models
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -210,8 +210,6 @@ https://opencode.ai/zen/v1/models
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** 表示価格には、2026年9月18日まで50%の割引が適用されています。
-
 **DeepSeek V4 Flash / Pro:** Peak時間は01:00-04:00と06:00-10:00 UTCで、それ以外の時間はすべてOff-Peakです。[詳しく見る](https://api-docs.deepseek.com/quick_start/pricing/)。

 使用履歴に Haiku、Nano、Flash などの[低コストモデル](/docs/config/#models)が表示されることがあります。OpenCode はこれらのモデルをセッションタイトルの生成に使用します。
@@ -184,8 +184,8 @@ https://opencode.ai/zen/v1/models
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -210,8 +210,6 @@ https://opencode.ai/zen/v1/models
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** 표시된 가격에는 2026년 9월 18일까지 50% 할인이 적용됩니다.
-
 **DeepSeek V4 Flash / Pro:** Peak 시간은 01:00-04:00 및 06:00-10:00 UTC이며, 그 외 모든 시간은 Off-Peak입니다. [자세히 알아보기](https://api-docs.deepseek.com/quick_start/pricing/).

 사용 기록에서 Haiku, Nano 또는 Flash와 같은 [저비용 모델](/docs/config/#models)을 볼 수 있습니다. OpenCode는 이러한 모델을 사용해 세션 제목을 생성합니다.
@@ -195,8 +195,8 @@ Vi støtter en pay-as-you-go-modell. Nedenfor er prisene **per 1M tokens**.
 | Grok 4.5 (> 200K tokens)          | $4.00   | $12.00  | $0.60         | -               |
 | Grok Build 0.1                    | $1.00   | $2.00   | $0.20         | -               |
 | Muse Spark 1.2                    | $1.25   | $4.25   | $0.15         | -               |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00   | $10.00  | $0.20         | $2.50           |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00   | $15.00  | $0.40         | $5.00           |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00   | $30.00  | $0.50         | $6.25           |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00  | $45.00  | $1.00         | $12.50          |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00   | $12.00  | $0.20         | $2.50           |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00   | $18.00  | $0.40         | $5.00           |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20   | $1.20   | $0.02         | $0.25           |
@@ -221,8 +221,6 @@ Vi støtter en pay-as-you-go-modell. Nedenfor er prisene **per 1M tokens**.
 | GPT 5 Codex                       | $1.07   | $8.50   | $0.107        | -               |
 | GPT 5 Nano                        | $0.05   | $0.40   | $0.005        | -               |

-**GPT 5.6 Sol:** Prisene som vises inkluderer 50 % rabatt til og med 18. september 2026.
-
 **DeepSeek V4 Flash / Pro:** Peak-tidene er 01:00-04:00 og 06:00-10:00 UTC; alle andre tider er Off-Peak. [Les mer](https://api-docs.deepseek.com/quick_start/pricing/).

 Du vil kanskje legge merke til [lavprismodeller](/docs/config/#models), som Haiku, Nano eller Flash, i brukshistorikken din. OpenCode bruker disse modellene til å generere økttitler.
@@ -195,8 +195,8 @@ Obsługujemy model pay-as-you-go. Poniżej znajdują się ceny **za 1M tokenów*
 | Grok 4.5 (> 200K tokens)          | $4.00   | $12.00  | $0.60          | -              |
 | Grok Build 0.1                    | $1.00   | $2.00   | $0.20          | -              |
 | Muse Spark 1.2                    | $1.25   | $4.25   | $0.15          | -              |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00   | $10.00  | $0.20          | $2.50          |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00   | $15.00  | $0.40          | $5.00          |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00   | $30.00  | $0.50          | $6.25          |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00  | $45.00  | $1.00          | $12.50         |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00   | $12.00  | $0.20          | $2.50          |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00   | $18.00  | $0.40          | $5.00          |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20   | $1.20   | $0.02          | $0.25          |
@@ -221,8 +221,6 @@ Obsługujemy model pay-as-you-go. Poniżej znajdują się ceny **za 1M tokenów*
 | GPT 5 Codex                       | $1.07   | $8.50   | $0.107         | -              |
 | GPT 5 Nano                        | $0.05   | $0.40   | $0.005         | -              |

-**GPT 5.6 Sol:** Podane ceny obejmują 50% zniżki do 18 września 2026 r.
-
 **DeepSeek V4 Flash / Pro:** Godziny Peak to 01:00-04:00 i 06:00-10:00 UTC; wszystkie pozostałe godziny to Off-Peak. [Dowiedz się więcej](https://api-docs.deepseek.com/quick_start/pricing/).

 W historii użycia możesz zauważyć [niedrogie modele](/docs/config/#models), takie jak Haiku, Nano lub Flash. OpenCode używa tych modeli do generowania tytułów sesji.
@@ -184,8 +184,8 @@ Oferecemos um modelo pay-as-you-go. Abaixo estão os preços **por 1M tokens**.
 | Grok 4.5 (> 200K tokens)          | $4.00   | $12.00  | $0.60            | -                |
 | Grok Build 0.1                    | $1.00   | $2.00   | $0.20            | -                |
 | Muse Spark 1.2                    | $1.25   | $4.25   | $0.15            | -                |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00   | $10.00  | $0.20            | $2.50            |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00   | $15.00  | $0.40            | $5.00            |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00   | $30.00  | $0.50            | $6.25            |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00  | $45.00  | $1.00            | $12.50           |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00   | $12.00  | $0.20            | $2.50            |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00   | $18.00  | $0.40            | $5.00            |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20   | $1.20   | $0.02            | $0.25            |
@@ -210,8 +210,6 @@ Oferecemos um modelo pay-as-you-go. Abaixo estão os preços **por 1M tokens**.
 | GPT 5 Codex                       | $1.07   | $8.50   | $0.107           | -                |
 | GPT 5 Nano                        | $0.05   | $0.40   | $0.005           | -                |

-**GPT 5.6 Sol:** Os preços exibidos incluem 50% de desconto até 18 de setembro de 2026.
-
 **DeepSeek V4 Flash / Pro:** Os horários Peak são 01:00-04:00 e 06:00-10:00 UTC; todos os demais horários são Off-Peak. [Saiba mais](https://api-docs.deepseek.com/quick_start/pricing/).

 Você pode notar [modelos de baixo custo](/docs/config/#models), como Haiku, Nano ou Flash, no seu histórico de uso. O OpenCode usa esses modelos para gerar títulos de sessões.
@@ -195,8 +195,8 @@ https://opencode.ai/zen/v1/models
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -221,8 +221,6 @@ https://opencode.ai/zen/v1/models
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** Указанные цены включают скидку 50% до 18 сентября 2026 года.
-
 **DeepSeek V4 Flash / Pro:** Часы Peak: 01:00-04:00 и 06:00-10:00 UTC; все остальные часы относятся к Off-Peak. [Подробнее](https://api-docs.deepseek.com/quick_start/pricing/).

 В истории использования могут появляться [недорогие модели](/docs/config/#models), такие как Haiku, Nano или Flash. OpenCode использует эти модели для создания заголовков сессий.
@@ -186,8 +186,8 @@ https://opencode.ai/zen/v1/models
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -212,8 +212,6 @@ https://opencode.ai/zen/v1/models
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** ราคาที่แสดงรวมส่วนลด 50% จนถึงวันที่ 18 กันยายน 2026
-
 **DeepSeek V4 Flash / Pro:** ช่วงเวลา Peak คือ 01:00-04:00 และ 06:00-10:00 UTC ส่วนเวลาอื่นทั้งหมดเป็น Off-Peak [ดูข้อมูลเพิ่มเติม](https://api-docs.deepseek.com/quick_start/pricing/)

 คุณอาจสังเกตเห็น[โมเดลต้นทุนต่ำ](/docs/config/#models) เช่น Haiku, Nano หรือ Flash ในประวัติการใช้งานของคุณ OpenCode ใช้โมเดลเหล่านี้เพื่อสร้างชื่อเซสชัน
@@ -184,8 +184,8 @@ Kullandıkça öde modelini destekliyoruz. Aşağıda **1M token başına** fiya
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -210,8 +210,6 @@ Kullandıkça öde modelini destekliyoruz. Aşağıda **1M token başına** fiya
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** Gösterilen fiyatlara 18 Eylül 2026 tarihine kadar %50 indirim dahildir.
-
 **DeepSeek V4 Flash / Pro:** Peak saatleri 01:00-04:00 ve 06:00-10:00 UTC'dir; diğer tüm saatler Off-Peak'tir. [Daha fazla bilgi](https://api-docs.deepseek.com/quick_start/pricing/).

 Kullanım geçmişinizde Haiku, Nano veya Flash gibi [düşük maliyetli modeller](/docs/config/#models) görebilirsiniz. OpenCode, oturum başlıklarını oluşturmak için bu modelleri kullanır.
@@ -195,8 +195,8 @@ We support a pay-as-you-go model. Below are the prices **per 1M tokens**.
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60       | -            |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20       | -            |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15       | -            |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20       | $2.50        |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40       | $5.00        |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50       | $6.25        |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00       | $12.50       |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20       | $2.50        |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40       | $5.00        |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02       | $0.25        |
@@ -221,8 +221,6 @@ We support a pay-as-you-go model. Below are the prices **per 1M tokens**.
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107      | -            |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005      | -            |

-**GPT 5.6 Sol:** Prices shown include a 50% discount through September 18, 2026.
-
 **DeepSeek V4 Flash / Pro:** Peak hours are 01:00-04:00 and 06:00-10:00 UTC; all other hours are Off-Peak. [Learn more](https://api-docs.deepseek.com/quick_start/pricing/).

 You may notice [low-cost models](/docs/config/#models), such as Haiku, Nano, or Flash, in your usage history. OpenCode uses these models to generate session titles.
@@ -184,8 +184,8 @@ https://opencode.ai/zen/v1/models
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60    | -        |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20    | -        |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15    | -        |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20    | $2.50    |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40    | $5.00    |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50    | $6.25    |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00    | $12.50   |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20    | $2.50    |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40    | $5.00    |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02    | $0.25    |
@@ -210,8 +210,6 @@ https://opencode.ai/zen/v1/models
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107   | -        |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005   | -        |

-**GPT 5.6 Sol:** 所示价格已包含 50% 折扣，有效期至 2026 年 9 月 18 日。
-
 **DeepSeek V4 Flash / Pro:** Peak 时段为 01:00-04:00 和 06:00-10:00 UTC；其他所有时段均为 Off-Peak。[了解更多](https://api-docs.deepseek.com/quick_start/pricing/)。

 你可能会在使用记录中看到 Haiku、Nano 或 Flash 等[低成本模型](/docs/config/#models)。OpenCode 使用这些模型生成会话标题。
@@ -189,8 +189,8 @@ https://opencode.ai/zen/v1/models
 | Grok 4.5 (> 200K tokens)          | $4.00  | $12.00  | $0.60    | -        |
 | Grok Build 0.1                    | $1.00  | $2.00   | $0.20    | -        |
 | Muse Spark 1.2                    | $1.25  | $4.25   | $0.15    | -        |
-| GPT 5.6 Sol (≤ 272K tokens)       | $2.00  | $10.00  | $0.20    | $2.50    |
-| GPT 5.6 Sol (> 272K tokens)       | $4.00  | $15.00  | $0.40    | $5.00    |
+| GPT 5.6 Sol (≤ 272K tokens)       | $5.00  | $30.00  | $0.50    | $6.25    |
+| GPT 5.6 Sol (> 272K tokens)       | $10.00 | $45.00  | $1.00    | $12.50   |
 | GPT 5.6 Terra (≤ 272K tokens)     | $2.00  | $12.00  | $0.20    | $2.50    |
 | GPT 5.6 Terra (> 272K tokens)     | $4.00  | $18.00  | $0.40    | $5.00    |
 | GPT 5.6 Luna (≤ 272K tokens)      | $0.20  | $1.20   | $0.02    | $0.25    |
@@ -215,8 +215,6 @@ https://opencode.ai/zen/v1/models
 | GPT 5 Codex                       | $1.07  | $8.50   | $0.107   | -        |
 | GPT 5 Nano                        | $0.05  | $0.40   | $0.005   | -        |

-**GPT 5.6 Sol:** 顯示價格已包含 50% 折扣，有效期至 2026 年 9 月 18 日。
-
 **DeepSeek V4 Flash / Pro:** Peak 時段為 01:00-04:00 和 06:00-10:00 UTC；其他所有時段均為 Off-Peak。[了解更多](https://api-docs.deepseek.com/quick_start/pricing/)。

 你可能會在使用紀錄中看到 Haiku、Nano 或 Flash 等[低成本模型](/docs/config/#models)。OpenCode 使用這些模型產生工作階段標題。