Problem#
maxOutput is null on every Grok chat row and every Mistral row.
| Provider | Rows without maxOutput | Examples |
|---|
| grok | 8 / 8 chat | grok-4.7, grok-4.6, grok-4.5, grok-build-0.1 |
| mistral | 52 / 52 | codestral-2508 |
| openai | 5 / 106 chat | gpt-5-search-api, gpt-live-1 |
GET /v1/models/grok/grok-4.7 already reports the gap in discrepancies:
{ "field": "maxOutput", "ours": null, "openrouter": 450000, "oursDerivation": null }
The same row also lists reasoning, include_reasoning, and top_k as ours: false, openrouter: true. grok-4.7 is a reasoning model.
Ask#
- Fill
maxOutput for Grok and Mistral from a native source only (listing, docs, or schema max_tokens.maximum). Never copy OpenRouter values into another provider's catalog: OpenRouter is a discrepancy signal only. If nothing native exists, leave the field null.
- Fill
capabilities.reasoning for grok-4.7. Check the other Grok chat rows too.
Acceptance#
- Every Grok chat row has
capabilities.reasoning set from native docs where xAI states it.
grok-4.7 discrepancies no longer lists capabilities.reasoning.
maxOutput is filled only where a native source states it. As of 2026-09-26 neither xAI nor Mistral publishes one: xAI's spec states only a 128k default, and Mistral caps output by the context length. So these rows stay null and the maxOutput discrepancy remains.