Greptile Summary

This PR fixes HTTP 422 rejections from the Mistral API by correcting maxTokens values across model definitions — the root cause was that several models had maxTokens set equal to their full contextWindow size, but max_tokens in the API controls output length, not input capacity, and Mistral enforces a much lower output limit.

• extensions/mistral/model-definitions.ts: MISTRAL_DEFAULT_MAX_TOKENS and five per-model maxTokens values are lowered to reasonable output limits (e.g. 16384 for mistral-large-latest, 32768 for devstral-medium-latest). The fix is correct for this module.
• src/commands/onboard-auth.test.ts: Test updated correctly to reflect the new mistral-large-latest value.
• Missing: extensions/mistral/model-definitions.test.ts was not updated — the assertions for magistral-small and pixtral-large-latest still expect the old value of 128000, so these tests will fail.
• Missing: src/plugins/provider-model-definitions.ts contains its own duplicate MISTRAL_DEFAULT_MAX_TOKENS = 262144 constant (line 30) used by a separate buildMistralModelDefinition() function, which was not updated. This leaves the original 422 bug alive for that code path.

Confidence Score: 2/5

• Not safe to merge — the fix is incomplete and existing tests will fail.
• Two concrete blockers prevent merging: (1) extensions/mistral/model-definitions.test.ts has stale assertions that will cause CI failures, and (2) a duplicate MISTRAL_DEFAULT_MAX_TOKENS = 262144 in src/plugins/provider-model-definitions.ts means the 422 bug is not fully resolved. Both are straightforward one-line fixes, but they need to be addressed before this PR is merged.
• extensions/mistral/model-definitions.test.ts (stale maxTokens assertions) and src/plugins/provider-model-definitions.ts (un-updated duplicate constant)

This is a comment left during a code review.
Path: extensions/mistral/model-definitions.test.ts
Line: 37-50

Comment:
**Stale test assertions will cause test failures**

The `maxTokens` values for `magistral-small` and `pixtral-large-latest` in this test still reflect the old values (`128000`), but the model definitions were updated to `40960` and `32768` respectively. These assertions will now fail.

```suggestion
        expect.objectContaining({
          id: "magistral-small",
          reasoning: true,
          input: ["text"],
          contextWindow: 128000,
          maxTokens: 40960,
        }),
        expect.objectContaining({
          id: "pixtral-large-latest",
          input: ["text", "image"],
          contextWindow: 128000,
          maxTokens: 32768,
        }),
```

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: src/plugins/provider-model-definitions.ts
Line: 30

Comment:
**Duplicate Mistral constant not updated**

This file contains its own copy of `MISTRAL_DEFAULT_MAX_TOKENS` that is still set to `262144`. The `buildMistralModelDefinition()` function at line 167 returns a model using this stale value, meaning the same 422 rejection bug persists for any code path that goes through this plugin-level builder rather than the `extensions/mistral` module.

```suggestion
const MISTRAL_DEFAULT_MAX_TOKENS = 16384;
```

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "fix(mistral): lower default maxTokens to..." | Re-trigger Greptile}

closed in favour of #53006