OpenAI-compatible gateway currently serving 13 free models. Free model
requests are rate limited to 200 requests/hour per IP, and the docs are
explicit that this applies to anonymous and authenticated requests alike,
so no account is needed.
Every free model reports mayTrainOnYourPrompts, which the section notes.
Based on #333, with the following corrections:
- The model name mappings did not match the ids the API actually returns
(poolside/laguna-xs.2:free vs poolside/laguna-xs-2.1:free), and included
ids that are not free (openrouter/owl-alpha) or not present at all
(nex-agi/nex-n2-pro:free), while missing three that are. Rebuilt the list
from the live isFree=true set. Since MODEL_TO_NAME_MAPPING is shared,
this also resolves nine raw ids in the OpenRouter section.
- Dropped the `or ":free" in model_id` fallback in the free-model filter;
isFree already covers every such id, and kilo-auto/free carries no suffix.
Closes#333
GitHub Models is being fully retired on July 30, 2026 -- the playground,
model catalogue, inference API and BYOK endpoints all go away, and it has
already been closed to new customers since June 16, 2026. It is no longer
a free API anyone can sign up for, so it does not belong in this list.
Removes the section, the fetch_github_models scraper and its wiring.
Reimplements #347 in src/pull_available_models.py: the original patch
edited the generated README.md directly, so the next scheduled run would
have reverted it.
https://github.blog/changelog/2026-07-01-github-models-is-being-fully-retired-on-july-30-2026/Closes#347
Verified every provider against its live API/docs. Updates the hardcoded
sections that nothing refreshes automatically:
- Cerebras: free tier is now gpt-oss-120b, zai-glm-4.7 and gemma-4-31b at
5 RPM / 30K TPM / 1M TPH / 1M TPD (was 30 RPM / 60K TPM / 14,400 RPD,
and still listed Llama 3.1 8B, which is gone).
- OpenCode Zen: corrected two model names and added the four free models
that were missing.
- Google AI Studio: Gemini 3.5 Flash was reading gemini-3-flash's quota.
Added Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, Gemma 4 31B and
Gemma 4 26B A4B, which all have free-tier quota.
- Mistral (La Plateforme): limits are set per-model and per-organization,
not a flat 1 RPS / 500K TPM. Verified against the admin console.
- Modal: the $5/month tier no longer exists; Starter is $30/month.
- Alibaba Model Studio: note the 90-day validity and Singapore-only endpoint.
- Scaleway: free tier also includes 60 minutes of audio transcription.
- Vercel AI Gateway: note that the free tier covers a subset of models.
Also fixes the dead Vertex AI block emitting an orphan </tbody></table>
into the README by guarding the closing tag with the same condition as
the opening one.
mistralai 2.x moved the Mistral class out of the top-level package,
breaking the daily CI run. The client was instantiated but never used,
so drop it along with the unused rate_limited_mistral_chat helper, and
pin remaining deps to avoid future surprise upgrades.
