Add catalog-only model discovery - #19
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: dfcfe8f685
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| provider: "anthropic", | ||
| model: "claude-fable-5", | ||
| pricing: { inputUsdPerMtok: 10, outputUsdPerMtok: 50 }, | ||
| contextWindow: 200000, |
There was a problem hiding this comment.
The Claude models overview linked from the docs shows Fable 5, Opus 4.8, and Sonnet 5 with 1M-token context windows (https://platform.claude.com/docs/en/about-claude/models/overview), and the existing candidates already use 1M for newer Opus/Sonnet models. Leaving these catalog entries at 200000 makes route models --catalog under-report long-context capacity by 5x, so users relying on the discovery metadata may incorrectly rule them out for prompts above 200k tokens.
Useful? React with 👍 / 👎.
|
|
||
| import { getDefaultCandidates, type ModelCandidate } from "@routerlab/core"; | ||
| import { | ||
| getCatalogModels, |
There was a problem hiding this comment.
This new import requires a symbol that does not exist in the currently allowed core range: packages/cli/package.json still permits @routerlab/core@^1.0.2, and changeset version only updates internal dependencies that are out of range, so the current range can stay satisfied by core 1.1.x. In installs or lockfiles that upgrade the CLI while keeping that older core version, the ESM import fails at module load before route models can run, so the dependency range needs to be raised with this API use.
Useful? React with 👍 / 👎.
| provider: "openai", | ||
| model: "gpt-5.5", | ||
| pricing: { inputUsdPerMtok: 5, outputUsdPerMtok: 30 }, | ||
| contextWindow: 400000, |
There was a problem hiding this comment.
The OpenAI model pages list gpt-5.5 and gpt-5.4 with 1,050,000-token context windows, and gpt-5.4-mini with a 400,000-token window (for example, https://developers.openai.com/api/docs/models/gpt-5.5 and https://developers.openai.com/api/docs/models/gpt-5.4-mini). These lower catalog values make route models --catalog under-report OpenAI long-context capacity, so users inspecting discovery metadata may incorrectly reject prompts above 400k for GPT-5.5/GPT-5.4 or above 200k for GPT-5.4 mini.
Useful? React with 👍 / 👎.
1 participant
Adds catalog-only model discovery for RouterLab without changing the default routed candidate pool.
This is the release PR that should trigger the changesets version workflow once merged.