Align ZAI modality metadata with source conventions

src/backend/drivers/ai-chat/providers/zai/ZAIProvider.test.ts

@@ -51,6 +51,10 @@ import { AIChatStream } from '../../utils/Streaming.js';
 import { ZAI_MODELS } from './models.js';
 import { ZAIProvider } from './ZAIProvider.js';
 import { modelLookupNames } from '../../utils/modelRouting.js';
+import {
+    modelSupportsModality,
+    requiredInputModalities,
+} from '../../utils/mediaParts.js';
 
 // ── OpenAI SDK mock ─────────────────────────────────────────────────
 //
@@ -211,6 +215,47 @@ describe('ZAIProvider model catalog', () => {
             cached_tokens: 7.5,
         });
     });
+
+    it('advertises image input, and no video or file input, only on the GLM-5.3 flash models', async () => {
+        const { provider } = makeProvider();
+        const models = await provider.models();
+
+        const imageIds = models
+            .filter((m) => modelSupportsModality(m, 'image'))
+            .map((m) => m.id);
+        expect(imageIds).toEqual(['glm-5.3-flash', 'glm-5.3-flashx']);
+        for (const m of models) {
+            expect(m.modalities?.output).toEqual(['text']);
+            expect(modelSupportsModality(m, 'video')).toBe(false);
+            expect(modelSupportsModality(m, 'file')).toBe(false);
+        }
+    });
+
+    it('passes the driver input gate for a puter.js image message on the flash models only', async () => {
+        const { provider } = makeProvider();
+        const models = await provider.models();
+        // The SDK image overload omits an explicit part type.
+        const required = requiredInputModalities([
+            {
+                content: [
+                    'what is this?',
+                    { image_url: { url: 'https://cdn.test/cat.png' } },
+                ],
+            },
+        ]);
+        expect(required).toEqual(['image']);
+
+        const accepts = (id: string) =>
+            required.every((modality) =>
+                modelSupportsModality(
+                    models.find((m) => m.id === id)!,
+                    modality,
+                ),
+            );
+        expect(accepts('glm-5.3-flash')).toBe(true);
+        expect(accepts('glm-5.3-flashx')).toBe(true);
+        expect(accepts('glm-5.3')).toBe(false);
+    });
 });
 
 // ── Request shape (OpenAI-compat quirks specific to GLM) ────────────
@@ -348,6 +393,40 @@ describe('ZAIProvider.complete request shape', () => {
         expect(args.response_format).toEqual({ type: 'json_object' });
     });
 
+    it.each(['glm-5.3-flash', 'zai/glm-5.3-flashx'])(
+        'forwards image_url parts for %s unchanged, by URL and by data URL',
+        async (requested) => {
+            const { provider } = makeProvider();
+            createMock.mockResolvedValueOnce(baseCompletion);
+            const content = [
+                { type: 'text', text: 'compare these' },
+                {
+                    type: 'image_url',
+                    image_url: { url: 'https://cdn.test/cat.png' },
+                },
+                {
+                    type: 'image_url',
+                    image_url: { url: 'data:image/png;base64,iVBORw0KGgo=' },
+                },
+            ];
+
+            await withTestActor(() =>
+                provider.complete({
+                    model: requested,
+                    messages: [
+                        { role: 'user', content: structuredClone(content) },
+                    ],
+                }),
+            );
+
+            const [args] = createMock.mock.calls[0]!;
+            expect(args.model).toBe(requested.replace(/^zai\//, ''));
+            // Z.AI reads http(s) and data URLs itself, so neither is fetched or
+            // rewritten on the way through.
+            expect(args.messages).toEqual([{ role: 'user', content }]);
+        },
+    );
+
     it('strips Anthropic-style cache_control from messages before sending', async () => {
         const { provider } = makeProvider();
         createMock.mockResolvedValueOnce(baseCompletion);

src/backend/drivers/ai-chat/providers/zai/models.ts

@@ -22,18 +22,22 @@ import { usdPerMToken } from '../../utils/pricing.js';
 
 const K = 1_000;
 
+// Advertise only media types handled by the chat adapter.
+const imageInput = ['text', 'image'];
+
 const textModel = (
     id: string,
     name: string,
     context: number,
     maxTokens: number,
     costs: IChatModel['costs'],
+    input: string[] = ['text'],
 ): IChatModel => ({
     puterId: `z-ai:z-ai/${id}`,
     id,
     name,
     aliases: [`z-ai/${id}`, `zai/${id}`],
-    modalities: { input: ['text'], output: ['text'] },
+    modalities: { input, output: ['text'] },
     open_weights: false,
     tool_call: true,
     context,
@@ -62,6 +66,7 @@ export const ZAI_MODELS: IChatModel[] = [
         1_000 * K,
         128 * K,
         usdPerMToken(0.15, 0.5, 0.03),
+        imageInput,
     ),
     textModel(
         'glm-5.3-flashx',
@@ -69,6 +74,7 @@ export const ZAI_MODELS: IChatModel[] = [
         1_000 * K,
         128 * K,
         usdPerMToken(0.37, 1.25, 0.075),
+        imageInput,
     ),
     textModel(
         'glm-5.2',

src/backend/drivers/ai-speech2txt/SpeechToTextDriver.test.ts

@@ -522,6 +522,24 @@ describe('SpeechToTextDriver model selection and payload shape', () => {
         expect(sent.timestamp_granularities).toBeUndefined();
     });
 
+    it('accepts but does not forward chunking or known-speaker options', async () => {
+        const { actor } = await makeUser();
+        transcriptionsCreateMock.mockResolvedValueOnce({ text: 'x' });
+
+        const result = await withActor(actor, () =>
+            driver.transcribe({
+                file: dataUrl(Buffer.from('a'), 'audio/mp3'),
+                chunking_strategy: 'auto',
+                known_speaker_names: ['agent'],
+            }),
+        );
+
+        expect(result).toEqual({ text: 'x' });
+        const sent = transcriptionsCreateMock.mock.calls[0]![0];
+        expect(sent.chunking_strategy).toBeUndefined();
+        expect(sent.known_speaker_names).toBeUndefined();
+    });
+
     it('passes extra_body through verbatim', async () => {
         const { actor } = await makeUser();
         transcriptionsCreateMock.mockResolvedValueOnce({ text: 'x' });

src/backend/drivers/ai-speech2txt/providers/xai/XAISpeechToTextProvider.test.ts

@@ -441,6 +441,31 @@ describe('XAISpeechToTextProvider request shape', () => {
         expect(form.get('model')).toBeNull();
     });
 
+    it('accepts but ignores legacy transcription options', async () => {
+        const { actor } = await makeUser();
+        fetchSpy.mockResolvedValueOnce(
+            sttResponse({ text: 'ok', duration: 1 }),
+        );
+
+        const result = await withActor(actor, () =>
+            driver.transcribe({
+                file: dataUrl(Buffer.from('a'), 'audio/mp3'),
+                timestamp_granularities: ['word'],
+                chunking_strategy: 'auto',
+                known_speaker_names: ['agent'],
+            }),
+        );
+
+        expect(result).toMatchObject({ text: 'ok' });
+        const form = (fetchSpy.mock.calls[0]![1] as RequestInit)
+            .body as FormData;
+        expect(form.get('timestamp_granularities')).toBeNull();
+        expect(form.get('timestamp_granularities[]')).toBeNull();
+        expect(form.get('chunking_strategy')).toBeNull();
+        expect(form.get('known_speaker_names')).toBeNull();
+        expect(form.get('known_speaker_names[]')).toBeNull();
+    });
+
     it('forwards a pinned grok-voice-transcribe version as `model`', async () => {
         const { actor } = await makeUser();
         fetchSpy.mockResolvedValueOnce(

src/backend/drivers/ai-speech2txt/types.ts

@@ -53,7 +53,6 @@ export interface ITranscribeArgs {
     timestamp_granularities?: string[];
     chunking_strategy?: string;
     known_speaker_names?: string[];
-    known_speaker_references?: unknown[];
     extra_body?: Record<string, unknown>;
     stream?: boolean;
     test_mode?: boolean;

src/backend/drivers/ai-tts/providers/elevenlabs/ElevenLabsTTSProvider.test.ts

@@ -218,6 +218,28 @@ describe('ElevenLabsTTSProvider catalog', () => {
         );
     });
 
+    it('prices each engine from the cost table in dollars per 1M chars', async () => {
+        const provider = makeProvider();
+        const engines = await provider.listEngines();
+        // toBe considers NaN equal to itself, so assert finiteness separately.
+        for (const cost of Object.values(ELEVENLABS_TTS_COSTS)) {
+            expect(Number.isFinite(cost)).toBe(true);
+        }
+        for (const engine of engines) {
+            expect(Number.isFinite(engine.pricing_per_million_chars)).toBe(
+                true,
+            );
+            expect(engine.pricing_per_million_chars).toBe(
+                ELEVENLABS_TTS_COSTS[engine.id] / 100,
+            );
+        }
+        const byId = Object.fromEntries(
+            engines.map((e) => [e.id, e.pricing_per_million_chars]),
+        );
+        expect(byId.eleven_multilingual_v2).toBe(162);
+        expect(byId.eleven_flash_v2_5).toBe(81);
+    });
+
     it.each(['eleven_turbo_v2_5', 'eleven_turbo_v2'])(
         'drops the deprecated %s from the listing',
         async (model) => {

src/backend/drivers/ai-tts/providers/elevenlabs/ElevenLabsTTSProvider.ts

@@ -271,7 +271,7 @@ export class ElevenLabsTTSProvider implements ITTSProvider {
             id: model.id,
             name: model.name,
             provider: 'elevenlabs',
-            pricing_per_million_chars: 0,
+            pricing_per_million_chars: ELEVENLABS_TTS_COSTS[model.id] / 100, // Convert microcents per character to dollars per million characters.
         }));
     }
 

src/backend/drivers/ai-tts/providers/xai/XAITTSProvider.test.ts

@@ -115,15 +115,18 @@ describe('XAITTSProvider catalog', () => {
         }
     });
 
-    it('listEngines reports a single xai-tts engine with pricing_per_million_chars', async () => {
+    it('listEngines reports a single xai-tts engine priced in dollars per 1M chars', async () => {
         const provider = makeProvider();
         const engines = await provider.listEngines();
         expect(engines).toHaveLength(1);
         expect(engines[0]).toMatchObject({
             id: 'xai-tts',
             provider: 'xai',
-            pricing_per_million_chars: 1500,
+            pricing_per_million_chars: 15,
         });
+        expect(engines[0].pricing_per_million_chars).toBe(
+            XAI_TTS_COSTS['xai-tts'] / 100,
+        );
     });
 });
 

src/backend/drivers/ai-tts/providers/xai/XAITTSProvider.ts

@@ -93,7 +93,7 @@ export class XAITTSProvider implements ITTSProvider {
                 id: 'xai-tts',
                 name: 'xAI TTS',
                 provider: 'xai',
-                pricing_per_million_chars: 1500,
+                pricing_per_million_chars: XAI_TTS_COSTS['xai-tts'] / 100, // Convert microcents per character to dollars per million characters.
             },
         ];
     }

src/docs/src/AI/txt2speech.listEngines.md

@@ -20,7 +20,7 @@ puter.ai.txt2speech.listEngines(options)
 
 A provider name to query. When passed as a string, this is shorthand for `{ provider }`. Defaults to `'aws-polly'`.
 
-Accepted values: `'aws-polly'`, `'openai'`, `'elevenlabs'`, `'gemini'`, `'xai'`, or `'all'` to list every provider at once.
+Accepted values are `'aws-polly'`, `'openai'`, `'elevenlabs'`, `'gemini'`, `'xai'`, `'speechify'`, or `'all'` to list every provider at once.
 
 Common aliases are also accepted (e.g. `'eleven'`, `'google'`, `'grok'`). An unrecognized provider is rejected with a `bad_request` error.
 
@@ -28,7 +28,7 @@ Common aliases are also accepted (e.g. `'eleven'`, `'google'`, `'grok'`). An unr
 
 | Option | Type | Description |
 |--------|------|-------------|
-| `provider` | `String` | TTS provider to query. Defaults to `'aws-polly'`; `'all'` returns every provider's engines |
+| `provider` | `String` | TTS provider to query. Defaults to `'aws-polly'`. The value `'all'` returns every provider's engines. |
 
 ## Return value
 

src/docs/src/AI/txt2speech.listVoices.md

@@ -19,7 +19,7 @@ puter.ai.txt2speech.listVoices(options)
 
 | Option | Type | Description |
 |--------|------|-------------|
-| `provider` | `String` | TTS provider to query. Defaults to `'aws-polly'`. Accepted: `'aws-polly'`, `'openai'`, `'elevenlabs'`, `'gemini'`, `'xai'`, or `'all'` to list every provider at once. Common aliases are also accepted (e.g. `'eleven'`, `'google'`, `'grok'`). |
+| `provider` | `String` | TTS provider to query. Defaults to `'aws-polly'`. Accepted values are `'aws-polly'`, `'openai'`, `'elevenlabs'`, `'gemini'`, `'xai'`, `'speechify'`, or `'all'` to list every provider at once. Common aliases are also accepted (e.g. `'eleven'`, `'google'`, `'grok'`). |
 | `engine` | `String` | Engine/model filter (provider-specific, ignored by some providers) |
 
 When `options` is a plain string it is treated as an `engine` filter for the default (AWS Polly) provider.

src/docs/src/Objects/ttsengine.md

@@ -21,4 +21,4 @@ The provider this engine belongs to, e.g. `'aws-polly'`, `'openai'`, `'elevenlab
 
 #### `pricing_per_million_chars` (Number)
 
-An optional cost per million characters. May be absent when the provider does not expose pricing.
+An optional cost in US dollars per million characters. May be absent when the provider does not expose pricing.

src/puter-js/src/modules/ai/ai.test.js

@@ -561,6 +561,20 @@ describe('ai.speech2txt driver payloads', () => {
         });
     });
 
+    it('speech2txt still sends options the backend accepts but ignores', async () => {
+        FakeXHR.respondWith = () => ({ success: true, result: { text: 'hi' } });
+        const legacy = {
+            timestamp_granularities: ['word'],
+            chunking_strategy: 'auto',
+            known_speaker_names: ['agent'],
+        };
+        await ai.speech2txt({ file: 'data:audio/mpeg;base64,QUJD', ...legacy });
+        expect(lastBody().args).toEqual({
+            file: 'data:audio/mpeg;base64,QUJD',
+            ...legacy,
+        });
+    });
+
     it('speech2txt returns bare text for response_format text', async () => {
         FakeXHR.respondWith = () => ({ success: true, result: { text: 'transcribed' } });
         const out = await ai.speech2txt({

src/puter-js/src/modules/ai/types.js

@@ -480,10 +480,9 @@
  * @property {boolean} [translate]
  * @property {number} [temperature]
  * @property {boolean} [logprobs]
- * @property {string[]} [timestamp_granularities]
- * @property {string} [chunking_strategy]
- * @property {string[]} [known_speaker_names]
- * @property {string[]} [known_speaker_references]
+ * @property {string[]} [timestamp_granularities] Accepted and ignored by all providers.
+ * @property {string} [chunking_strategy] Accepted and ignored by all providers.
+ * @property {string[]} [known_speaker_names] Accepted and ignored by all providers.
  * @property {Record<string, unknown>} [extra_body]
  * @property {boolean} [format]
  * @property {boolean} [diarize]

src/puter-js/tests/api/suites/ai.suite.ts

@@ -1300,6 +1300,18 @@ export default suite('ai', {
         t.assert.equal(error.message, 'Audio input is required');
     },
 
+    'speech2txt accepts ignored options in test mode': async (t) => {
+        useApiToken(t);
+        const result = (await t.puter.ai.speech2txt({
+            file: TINY_AUDIO,
+            timestamp_granularities: ['word'],
+            chunking_strategy: 'auto',
+            known_speaker_names: ['speaker'],
+            test_mode: true,
+        })) as { text?: string };
+        t.assert.ok((result.text ?? '').includes('sample transcription'));
+    },
+
     'speech2txt honors the positional test-mode flag': async (t) => {
         useApiToken(t);
         const result = (await t.puter.ai.speech2txt(