Repository navigation
[FEATURE]: Route @ai-sdk/xai models through the V2 native responses runner #35100
Description
Feature hasn't been suggested before.
- I have verified this feature I'm about to request hasn't been suggested before.
Describe the enhancement you want to request
The V2 session runner sets a prompt cache key on every request (packages/core/src/session/runner/llm.ts), and the OpenAI Responses protocol writes it to the prompt_cache_key body field. SessionRunnerModel.fromCatalogModel has no branch for @ai-sdk/xai, so xai catalog models fail with UnsupportedApiError and never reach the native route that carries the key.
Request: map @ai-sdk/xai catalog models onto the existing OpenAIResponses route. xAI's Responses API matches the OpenAI Responses API, including prompt_cache_key, so no new protocol work is needed. One detail: the route's default baseURL points at api.openai.com, so xai models without a catalog URL need a default of https://api.x.ai/v1.
Benefit: xai sessions on the V2 runner send the identifier that xAI uses to route requests to the server holding their cache. Cached tokens are billed at a reduced rate and return faster. See the xAI prompt caching docs.
Related: #35033 covers the same gap in the legacy provider path. This issue is specific to the V2 native runner. It replaces #35034, which was closed for not using the issue template.
Activity
opencode-agent commentedon Jul 8, 2026
Audit update (2026-07-08): #35031 is the active implementation, but it currently targets
devrather thanv2. Confirm the intended landing branch or port plan before treating this V2 issue as complete.opencode-agent commentedon Jul 30, 2026
Addressed by #39613 together with the native xAI provider package on
v2. xAI catalog models now resolve through the provider-package contract with credentials, base URL, headers, body overlays, limits, and provider options applied; the xAI provider defaultsmodel()to its native OpenAI Responses route and uses the xAI API base URL. This lets V2 catalog-selected xAI models carry Responses options such as the prompt cache key instead of failing package resolution. Closing as resolved.