fix(core): retain raw copilot completion usage

packages/core/src/github-copilot/chat/openai-compatible-chat-language-model.ts

@@ -285,9 +285,8 @@ export class OpenAICompatibleChatLanguageModel implements LanguageModelV3 {
           cacheWrite: undefined,
         },
         outputTokens: {
-          total: responseBody.usage?.completion_tokens ?? undefined,
+          ...outputUsage(responseBody.usage),
           text: undefined,
-          reasoning: responseBody.usage?.completion_tokens_details?.reasoning_tokens ?? undefined,
         },
         raw: responseBody.usage ?? undefined,
       },
@@ -357,6 +356,7 @@ export class OpenAICompatibleChatLanguageModel implements LanguageModelV3 {
         cachedTokens: number | undefined
       }
       totalTokens: number | undefined
+      rawCompletionTokens: number | undefined
     } = {
       completionTokens: undefined,
       completionTokensDetails: {
@@ -369,6 +369,7 @@ export class OpenAICompatibleChatLanguageModel implements LanguageModelV3 {
         cachedTokens: undefined,
       },
       totalTokens: undefined,
+      rawCompletionTokens: undefined,
     }
     let isFirstChunk = true
     const providerOptionsName = this.providerOptionsName
@@ -432,11 +433,11 @@ export class OpenAICompatibleChatLanguageModel implements LanguageModelV3 {
               } = value.usage
 
               usage.promptTokens = prompt_tokens ?? undefined
-              usage.completionTokens = completion_tokens ?? undefined
+              usage.rawCompletionTokens = completion_tokens ?? undefined
+              const output = outputUsage(value.usage)
+              usage.completionTokens = output.total
+              usage.completionTokensDetails.reasoningTokens = output.reasoning
               usage.totalTokens = total_tokens ?? undefined
-              if (completion_tokens_details?.reasoning_tokens != null) {
-                usage.completionTokensDetails.reasoningTokens = completion_tokens_details?.reasoning_tokens
-              }
               if (completion_tokens_details?.accepted_prediction_tokens != null) {
                 usage.completionTokensDetails.acceptedPredictionTokens =
                   completion_tokens_details?.accepted_prediction_tokens
@@ -708,7 +709,7 @@ export class OpenAICompatibleChatLanguageModel implements LanguageModelV3 {
                 },
                 raw: {
                   prompt_tokens: usage.promptTokens ?? null,
-                  completion_tokens: usage.completionTokens ?? null,
+                  completion_tokens: usage.rawCompletionTokens ?? null,
                   total_tokens: usage.totalTokens ?? null,
                 },
               },
@@ -727,6 +728,7 @@ const openaiCompatibleTokenUsageSchema = z
   .object({
     prompt_tokens: z.number().nullish(),
     completion_tokens: z.number().nullish(),
+    reasoning_tokens: z.number().nullish(),
     total_tokens: z.number().nullish(),
     prompt_tokens_details: z
       .object({
@@ -743,6 +745,17 @@ const openaiCompatibleTokenUsageSchema = z
   })
   .nullish()
 
+function outputUsage(usage: z.infer<typeof openaiCompatibleTokenUsageSchema>) {
+  const nested = usage?.completion_tokens_details?.reasoning_tokens
+  return {
+    total:
+      usage?.completion_tokens == null
+        ? undefined
+        : usage.completion_tokens + (nested == null ? (usage.reasoning_tokens ?? 0) : 0),
+    reasoning: nested ?? usage?.reasoning_tokens ?? undefined,
+  }
+}
+
 // limited version of the schema, focussed on what is needed for the implementation
 // this approach limits breakages when the API changes and increases efficiency
 const OpenAICompatibleChatResponseSchema = z.object({

packages/core/test/github-copilot/copilot-chat-model.test.ts

@@ -204,7 +204,7 @@ describe("doStream", () => {
       finishReason: { unified: "tool-calls" },
       usage: {
         inputTokens: { total: 19581 },
-        outputTokens: { total: 53 },
+        outputTokens: { total: 187, reasoning: 134 },
       },
     })
   })
@@ -259,7 +259,7 @@ describe("doStream", () => {
       finishReason: { unified: "stop" },
       usage: {
         inputTokens: { total: 5778 },
-        outputTokens: { total: 59 },
+        outputTokens: { total: 154, reasoning: 95 },
       },
       providerMetadata: {
         copilot: {
@@ -391,7 +391,7 @@ describe("doStream", () => {
       finishReason: { unified: "tool-calls" },
       usage: {
         inputTokens: { total: 3767 },
-        outputTokens: { total: 19 },
+        outputTokens: { total: 30, reasoning: 11 },
       },
     })
   })