Repository navigation
Clean docs against the live model catalogue for llms.txt - #725
Merged
Merged
Conversation
Fifteen pages documented live models under ids the API does not know: a doubled suffix from the page generator (kimi-k3-1-pro-highspeed-highspeed, seed-1.6-flash-flash, eleven_v4_turbo_turbo, ...), dashed ids for models the catalogue lists dotted (gpt-5.1-codex, gpt-5.2-codex), or another model's id copied in (image-01 listed google/imagen-4.0-generate-001). Each id now matches GET /v1/models, and dead snapshot ids are dropped from pages that keep a live one.
GET /v1/models was taken as the source of truth on 2026-11-14. 140 model pages (76 of them hidden) documented ids that are withdrawn or past their shutdown date, plus 25 legacy swagger-era pages that SUMMARY.md no longer listed (old kling/, minimax-video/, speech-voice-models/ and runway files). They are deleted with the OpenAPI specs only they used, and their SUMMARY entries, list items, table rows and content-ref blocks go with them. model-database.md: 83 rows for ids missing from the catalogue move into the Deprecated table, 6 ids that only had a page are added there, deprecated rows stop linking to deleted pages, and 19 live models (HeyGen, devstral-2512, qwen-max, ...) that sat in the Deprecated table go back to their sections. The Kling AI overview now lists the models that remain.
261 JSON files under api-references/ were not referenced by any page, SUMMARY or GitBook config: specs of pages removed long ago, *-test-copy and *_local_changes leftovers, and duplicates of specs that moved. No page renders them, so docs.aimlapi.com does not change. If a spec with one of these names is still registered in GitBook's OpenAPI section, its refresh will 404 and it can be deleted there.
gpt-6.1-terra and mistral-large-4-0 were generated from the GLM 5.3 FlashX page and still listed z-ai/glm-5.3-flashx in the model hint, linked its Playground and showed provider zhipu; their specs and code were already right. LongCat Distilled 480p was a copy of the LTXV 2 page under another name, and no LongCat video model exists, so it goes. Also removes three unlisted pages with dead links: the Bagoodex search overview, the old vision-models table and a duplicate of the 404 page.
scripts/check-catalogue.py fails when a model page lists an id that GET /v1/models does not serve or when a relative .md link is broken, and warns when two model pages list the same id under unrelated names (a copied page) or when a catalogue docsUrl has no page here. Runs on docs PRs and daily, since a model can be withdrawn without a change to this repo. On main before this branch it reports 170 errors; on this branch none.
GitBook prints a page's description next to its link in /llms.txt, and 857 of 906 links there had none, so an agent reading the index saw eight links titled "OpenAI" and model pages whose slug does not match the id (claude-sonnet-5.5 serves anthropic/claude-sonnet-5-5). Model pages now start their description with the model id followed by the catalogue's own summary of the model; provider overviews say which task and provider they cover; the quickstart, SDK, MCP, Claude Code and service endpoint pages say what an agent needs to call them (base URL, endpoint, whether a key is needed). The Quickstart no longer advertises GPT-4o, and the hidden page titled llms.txt, which pointed at a dead OpenAPI URL, is removed.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Agents reading
docs.aimlapi.com/llms.txtandllms-full.txt(generated by GitBook from this repo) hit withdrawn models, mistyped ids and copied pages. This PR makes the docs matchGET /v1/models(816 live ids on 2026-11-14) and gives the llms.txt index useful descriptions.kimi-k3-1-pro-highspeed-highspeed,eleven_v4_turbo_turbo, …), dashed ids the catalogue lists dotted (gpt-5.1-codex), another model's id copied in (image-01).llms-full.txt) and 25 legacy swagger-era pages absent fromSUMMARY.md. SUMMARY entries, list items, table rows and content-ref blocks that pointed at them go too. Pages that a live model's cataloguedocsUrlpoints at were kept and re-pointed.devstral-2512,qwen-max, …) moved back out of it.gpt-6.1-terraandmistral-large-4-0listedz-ai/glm-5.3-flashx; LongCat Distilled 480p was a copy of LTXV 2 and is removed.Model id: <id>.plus the catalogue summary; provider overviews and key pages (quickstart, SDKs, MCP, Claude Code, service endpoints) say what an agent needs. The hidden page titled "llms.txt" (dead OpenAPI URL) is removed.scripts/check-catalogue.py+.github/workflows/check-catalogue.yml(PRs and daily). Onmainbefore this branch: 170 errors. On this branch: 0 errors, 27 warnings (cataloguedocsUrlpointing at pages that do not exist — a backend model-card fix).Verification
python3 scripts/check-catalogue.py→0 error(s), 27 warning(s); the same script onorigin/main→ 170 errors.description..mdlink indocs/points at a missing file.