Problem#
#73 added pricing sources. Some chat rows still have pricing: null.
Measured on GET /v1/models?pricing=1 on 2026-10-03:
| Provider | Chat rows without a card | Examples |
|---|
| together | 112 / 187 | agentica-org/DeepCoder-14B-Preview, allenai/Molmo-7B-D-0924 |
| cohere | 17 / 17 | command-a-03-2025, command-a-plus-05-2026 |
| gemini | 16 / 43 | antigravity-preview-09-2026, aqa |
| mistral | 15 / 33 | magistral-medium-latest, glm-5-2, labs-leanstral-1-5 |
| openai | 4 / 107 | gpt-5-search-api, gpt-3.5-turbo-16k |
| byteplus | 2 / 13 | deepseek-v4-1-flash-260910 |
Also, no Groq card (6) and no Together card (75) has a cache_read_tokens rate.
Ask#
- Add a pricing source for Cohere.
- Find why 112 Together chat rows have no card when 75 do.
- Fill the Mistral and Gemini rows where the provider's price page lists the model. Leave preview models with no published price as null.
- Check whether Groq and Together publish a cached-input price. If they do, add it.
Acceptance#
GET /v1/status shows a non-zero counts.priced for cohere.
GET /v1/models/mistral/magistral-medium-latest returns a rate card.
- Together
priced / models is above 90%, or the unpriced rows are ones Together does not price per token.