app: model tooltips show reasoning as unsupported despite reasoning variants #53520

Activity

1. github-actions

   assigned

   Brendonovich

   on Oct 6, 2026
2. github-actions

   added

   needs:complianceThis means the issue will auto-close after 2 hours.This means the issue will auto-close after 2 hours.

   on Oct 6, 2026
3. Oktinner0320 commented on Oct 6, 2026

   

   Oktinner0320

   on Oct 6, 2026

   Author

   More actions

   Now i improve
4. github-actions

   removed

   needs:complianceThis means the issue will auto-close after 2 hours.This means the issue will auto-close after 2 hours.

   on Oct 6, 2026
5. github-actions commented on Oct 6, 2026

   

   github-actionsBot

   on Oct 6, 2026 – with GitHub Actions

   Contributor

   More actions

   Thanks for updating the issue with the reproduction steps and system details!
6. yukunmeng commented on Oct 20, 2026

   

   yukunmeng

   on Oct 20, 2026

   More actions

   Confirmed on opencode 2.0.32 (Windows x64, Chinese UI): the tooltip shows 不支持推理 for every model in the picker, not only the ones named here.

   Measured against a live 2.0.32 server (GET /api/model):

   • 47 models returned
   • 0 / 47 have capabilities.reasoning
   • 0 / 47 have a top-level reasoning field
   • all 47 have capabilities present, so the model.reasoning fallback in the tooltip is unreachable

   Root cause, still present on current dev:

   • packages/core/src/models-dev.ts → modelInfo() builds capabilities from tools / input / output / temperature only. models.dev's top-level reasoning: true (and reasoning_options) is never copied into capabilities; it is only converted into variants by supports().
   • packages/app/src/providers/models/tooltip.tsx → reasoning() reads props.model.capabilities.reasoning (a boolean) when capabilities exists, and only falls back to props.model.reasoning otherwise. In V2 capabilities is always set, so the fallback is dead code.

   Reasoning itself works — only the label is wrong. From GET /api/session on the same server:

   • a session on opencode/space-bunny-free#max reports tokens.reasoning: 71263
   • a session on opencode/minimax-m3.1-flash-free#max reports tokens.reasoning: 16437
   • opencode run --model "wb-proxy-cn/space-bunny#high" completes normally with the #high variant applied

   So the badge should be derived from variants.length > 0, or the server should carry capabilities.reasoning through to the client.

   Two related findings, in case they are useful for the fix:

   1. A config entry replaces the catalog entry instead of merging with it. Adding

      "providers": { "opencode": { "models": { "space-bunny-free": { "capabilities": { "reasoning": { "supported": true, "effort": ["low","medium","high","xhigh","max"] } } } } } }

      to opencode.json did not merge with the models.dev record. The result was name: "space-bunny-free" (raw id), limit: {context: 200000, output: 32000} (unknown-model fallback), input: ["text","image"] (the video modality was lost) and compatibility.reasoningField was dropped. Related: Config.Model.Capabilities.reasoning in the config schema is the v2 object shape (supported/effort/budget/toggle) used to build variants, while the runtime Model.Capabilities type has no reasoning field at all — so it can never reach the client, and the v2 object shape would not satisfy the tooltip's boolean check even if it did.

   2. opencode/space-bunny-free is now status: "deprecated" in models.dev. After a catalog refresh it disappears from the picker and any run fails with Model space-bunny-free has been deprecated. Use minimax-m3.1-flash-free instead. Re-pinning it in opencode.json (spelling out limit / capabilities / compatibility / variants) makes it visible again with correct metadata, but the server still refuses to execute it.

   Environment: opencode v2.0.32, Windows 10.0.26300 (x64), UI language zh.
7. neriousy commented on Oct 20, 2026

   

   neriousy

   on Oct 20, 2026

   Member

   More actions

   Thanks for the detailed report and the GPT-6 examples. This is the same problem as #50257: reasoning works, but the model tooltip incorrectly says it is unsupported, and we're tracking it there. Please subscribe or add a 👍 to that issue, and share any further model or version details there.
8. neriousy

   closed this as a duplicateof desktop: model picker shows "No reasoning" for all V2 models — capabilities.reasoning is never set #50257 on Oct 20, 2026