Repository navigation
Mistral codestral-2508 has incorrect context window (256K → 128K) #745
Description
Summary
The codestral-2508 entry in the model catalog has max_input_tokens: 256000, but Mistral's official documentation lists the context window as 128K tokens. The parent entry codestral-latest already has the correct value of 128000.
Current vs Expected
| Field | Current | Expected | Source |
|---|---|---|---|
max_input_tokens |
256,000 | 128,000 | Mistral model card |
Affected entries
codestral-2508atpackages/proxy/schema/model_list.jsonline 5077
Notes
- The parent entry
codestral-latest(line 5059) already has the correctmax_input_tokens: 128000 - The sibling entry
codestral-2501(line 5070) inherits from the parent and is also correct - Only
codestral-2508explicitly overridesmax_input_tokensto the incorrect value of 256000
Verification
- Mistral model card: https://docs.mistral.ai/models/model-cards/codestral-25-08 — lists 128k context window
- Cross-check: Parent entry
codestral-latestin same catalog uses 128000
Activity
vinsymoyya commentedon Nov 3, 2026
Opened #1711 to correct
codestral-2508.max_input_tokensfrom 256,000 to 128,000, matching Mistral's current Codestral 25.08 model card.The change is limited to that field. The existing catalog test passes all 26 cases. All 1,471 entries also validate against the built schema, and the public model API returns the corrected value. A direct assertion fails before the fix and passes afterward; a parsed-JSON comparison confirms every other catalog value is unchanged.
git diff --checkpasses.Hosted CI currently requires maintainer approval (
action_required). The PR records the validation commands and the limitation that the complete package declaration build was interrupted; a full build/test-suite pass is not claimed.