@ai-sdk/gateway 4.0.90 → 4.0.91

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # @ai-sdk/gateway
 
+## 4.0.91
+
+### Patch Changes
+
+- 2693319: Add Gemini 3.8 TTS support with structured speech metadata and per-turn speaker and style controls for prebuilt voices. Preserve native WAV responses without adding a second header, support explicit raw PCM, mu-law, and A-law output, and identify headerless audio formats correctly. Add the Gemini 3.8 speech model IDs to Google and Gateway types.
+
+  Share transcript and custom-voice inspection through the Google provider internal export, and reject empty speech transcripts before sending a request. Default newer and custom model IDs to structured speech while preserving the legacy format for Gemini 2.5 and 3.1.
+
+- b73f2f9: feat(provider/gateway): add quantization conditions to the has provider option
+- Updated dependencies [fe07867]
+- Updated dependencies [a4b0940]
+- Updated dependencies [771e74b]
+  - @ai-sdk/provider-utils@5.0.47
+
 ## 4.0.90
 
 ### Patch Changes

package/dist/index.d.ts

@@ -81,7 +81,7 @@ type GatewayRealtimeModelId = 'google/gemini-3.8-live' | 'google/gemini-3.8-live
 
 type GatewayRerankingModelId = 'cohere/rerank-v3.5' | 'cohere/rerank-v4-fast' | 'cohere/rerank-v4-pro' | 'voyage/rerank-2.5' | 'voyage/rerank-2.5-lite' | (string & {});
 
-type GatewaySpeechModelId = 'fish-audio/s1' | 'fish-audio/s2-pro' | 'fish-audio/s2.1-pro' | 'openai/tts-1' | 'openai/tts-1-hd' | 'spacexai/grok-tts' | (string & {});
+type GatewaySpeechModelId = 'fish-audio/s1' | 'fish-audio/s2-pro' | 'fish-audio/s2.1-pro' | 'google/gemini-3.8-flash-tts' | 'google/gemini-3.8-flash-lite-tts' | 'openai/tts-1' | 'openai/tts-1-hd' | 'spacexai/grok-tts' | (string & {});
 
 type GatewayTranscriptionModelId = 'fish-audio/transcribe-1' | 'google/gemini-3.5-transcribe' | 'google/gemini-3.5-transcribe-live' | 'openai/gpt-4o-mini-transcribe' | 'openai/gpt-4o-transcribe' | 'openai/gpt-realtime-whisper' | 'openai/whisper-1' | 'spacexai/grok-stt' | (string & {});
 
@@ -1120,11 +1120,17 @@ type GatewayProviderOptions = {
     /** Filter to providers that do not train on prompt data. */
     disallowPromptTraining?: boolean;
     /**
-     * Restrict routing to models that have all of the given capabilities.
-     * Currently supports `'implicit-caching'`, `'reasoning'`, `'tool-use'`, and
-     * `'vision'` (image input).
-     */
-    has?: Array<'implicit-caching' | 'reasoning' | 'tool-use' | 'vision'>;
+     * Restrict routing to provider models that satisfy every given entry.
+     *
+     * Entries are capability tags (`'implicit-caching'`, `'reasoning'`,
+     * `'tool-use'`, `'vision'`) or weight-format conditions: `'quantization:fp8'`
+     * requires the serving provider to report that weight format,
+     * `'!quantization:fp8'` excludes it (providers with no recorded format still
+     * pass an exclusion). Format values are an open space but must match
+     * `[a-zA-Z0-9._-]{1,32}` and compare case-insensitively; unknown capability
+     * names are rejected by the Gateway with a 400.
+     */
+    has?: Array<'implicit-caching' | 'reasoning' | 'tool-use' | 'vision' | `quantization:${string}` | `!quantization:${string}`>;
     /**
      * Idempotency key for `experimental_startBatch`: retries with the same
      * key replay the original batch instead of creating a duplicate.

package/dist/index.js

@@ -3662,7 +3662,7 @@ async function getVercelRequestId() {
 }
 
 // src/version.ts
-var VERSION = true ? "4.0.90" : "0.0.0-test";
+var VERSION = true ? "4.0.91" : "0.0.0-test";
 
 // src/gateway-provider.ts
 var AI_GATEWAY_PROTOCOL_VERSION = "0.0.1";

package/docs/00-ai-gateway.mdx

@@ -1390,7 +1390,7 @@ The following gateway provider options are available:
 
   The unique identifier for the entity against which quota is tracked. Used for quota management and enforcement purposes.
 
-- **has** _Array<'implicit-caching' | 'reasoning' | 'tool-use' | 'vision'>_
+- **has** _Array<'implicit-caching' | 'reasoning' | 'tool-use' | 'vision' | `quantization:${string}` | `!quantization:${string}`>_
 
   Restricts routing to provider models that have all of the specified capabilities. Applies to both BYOK and system credentials, since the capability is a property of the model rather than the credential. If no provider model for the requested model satisfies the capabilities, the request fails. Unsupported values are rejected.
 
@@ -1400,7 +1400,9 @@ The following gateway provider options are available:
   - `'tool-use'` — models that support tool calling.
   - `'vision'` — models that accept image input.
 
-  Example: `has: ['implicit-caching']` will only route to models that support implicit caching. Example: `has: ['vision']` will only route to models that accept image input. Example: `has: ['tool-use']` will only route to models that support tool calling.
+  Weight-format conditions route on the serving provider's recorded weight format: `'quantization:fp8'` requires a provider serving fp8 weights, while `'!quantization:fp8'` excludes fp8 providers. A negated condition also matches providers whose weight format is not recorded, so exclusions remain usable while the catalog is populated.
+
+  Example: `has: ['implicit-caching']` will only route to models that support implicit caching. Example: `has: ['vision']` will only route to models that accept image input. Example: `has: ['tool-use']` will only route to models that support tool calling. Example: `has: ['!quantization:fp8']` will only route to providers that do not serve fp8 weights.
 
 - **providerTimeouts** _object_
 
@@ -1538,7 +1540,7 @@ const { text } = await generateText({
 
 #### Filtering by Model Capability
 
-Set `has` to restrict routing to provider models that have the specified capabilities. This applies to both BYOK and system credentials, since the capability is a property of the model rather than the credential. `'implicit-caching'` limits routing to models that perform automatic prompt caching, `'vision'` limits routing to models that accept image input, `'reasoning'` limits routing to models that support reasoning, and `'tool-use'` limits routing to models that support tool calling. If no provider model for the requested model satisfies the capabilities, the request fails.
+Set `has` to restrict routing to provider models that have the specified capabilities. This applies to both BYOK and system credentials, since the capability is a property of the model rather than the credential. `'implicit-caching'` limits routing to models that perform automatic prompt caching, `'vision'` limits routing to models that accept image input, `'reasoning'` limits routing to models that support reasoning, and `'tool-use'` limits routing to models that support tool calling. Weight-format conditions (`'quantization:fp8'` to require, `'!quantization:fp8'` to exclude) limit routing by the serving provider's recorded weight format. If no provider model for the requested model satisfies the capabilities, the request fails.
 
 ```ts
 import type { GatewayProviderOptions } from '@ai-sdk/gateway';

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@ai-sdk/gateway",
   "private": false,
-  "version": "4.0.90",
+  "version": "4.0.91",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,
@@ -31,7 +31,7 @@
   },
   "dependencies": {
     "@ai-sdk/provider": "4.0.18",
-    "@ai-sdk/provider-utils": "5.0.46",
+    "@ai-sdk/provider-utils": "5.0.47",
     "@vercel/oidc": "3.2.0"
   },
   "devDependencies": {

package/src/gateway-provider-options.ts

@@ -16,11 +16,24 @@ export type GatewayProviderOptions = {
   disallowPromptTraining?: boolean;
 
   /**
-   * Restrict routing to models that have all of the given capabilities.
-   * Currently supports `'implicit-caching'`, `'reasoning'`, `'tool-use'`, and
-   * `'vision'` (image input).
+   * Restrict routing to provider models that satisfy every given entry.
+   *
+   * Entries are capability tags (`'implicit-caching'`, `'reasoning'`,
+   * `'tool-use'`, `'vision'`) or weight-format conditions: `'quantization:fp8'`
+   * requires the serving provider to report that weight format,
+   * `'!quantization:fp8'` excludes it (providers with no recorded format still
+   * pass an exclusion). Format values are an open space but must match
+   * `[a-zA-Z0-9._-]{1,32}` and compare case-insensitively; unknown capability
+   * names are rejected by the Gateway with a 400.
    */
-  has?: Array<'implicit-caching' | 'reasoning' | 'tool-use' | 'vision'>;
+  has?: Array<
+    | 'implicit-caching'
+    | 'reasoning'
+    | 'tool-use'
+    | 'vision'
+    | `quantization:${string}`
+    | `!quantization:${string}`
+  >;
 
   /**
    * Idempotency key for `experimental_startBatch`: retries with the same

package/src/gateway-speech-model-settings.ts

@@ -2,6 +2,8 @@ export type GatewaySpeechModelId =
   | 'fish-audio/s1'
   | 'fish-audio/s2-pro'
   | 'fish-audio/s2.1-pro'
+  | 'google/gemini-3.8-flash-tts'
+  | 'google/gemini-3.8-flash-lite-tts'
   | 'openai/tts-1'
   | 'openai/tts-1-hd'
   | 'spacexai/grok-tts'