chore: backport known models catalog and AI price book to 2.37 (#28887, #28909, #28939, #28955, #29170, #29218, #29226, #29471, #29585) - #29588
Conversation
…amp its reasoning effort Backport of the chatd portion of #28955 (9cd66d3). GPT-6 Astra's function calling is Responses-only and the pinned SDK predates the model, so UsesResponsesAPI defaults it to Responses; it also rejects reasoning effort none and lists no minimal, so both clamp to low. (cherry picked from commit 9cd66d3, chatd files only)
…/2.37 Bring scripts/aibridgepricesgen/curation.json and overrides.jq to their main state and regenerate prices.json and knownModelsGenerated.json with make gen/aibridge-prices on this branch. This carries the model and price changes from #28887, #28909, #28939, #28955, #29170, #29218, #29226, #29471 and #29585 without cherry-picking the regenerated artifacts across the modules/aiModels move (#29040), which is not backported. The Anthropic thinking-mode pin gains the same entries as on main.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
👋 Hey @ibetitsmike! This PR is targeting the Only bug fixes should be cherry-picked to release branches. If this is a bug fix, please update the PR title to match the conventional commit format: If this is not a bug fix, it likely should not target a release branch. |
![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: 7939e2a76e
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".
…TURE.md They were placeholders for the human author of #28955 and main has since replaced them with real prose; the backport should not carry them into the release branch.
The fantasy pin on release/2.37 predates gpt-6-astra, and the Azure provider exposes no hook to force Responses, so an Azure Astra model config would speak Chat Completions while Astra's function calling is Responses-only. main added the Azure entry after bumping fantasy (#29230), which the backport does not carry. TestCuratedAstraUsesResponses guards the curation against reintroducing an Astra entry whose resolved transport is not Responses.
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: fd0e070427
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".
|
@codex review |
|
Codex Review: Didn't find any major issues. Hooray! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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". |
2 participants
Backport of the known models catalog and AI Gateway price book updates that landed on main since the August 27 refresh (#28720 was the last one on this branch), so 2.37 suggests the current models in
/agentssettings and prices them correctly in AI Bridge.Source PRs: #28887 (Claude Fable 5.1, Opus 5, Gemini 3.7 Flash, GLM 5.3), #28909 (Gemini 3.8 Flash), #28939 and #29471 (price book refreshes), #28955 (GPT-6 Astra), #29170 (gpt-6-astra and qwen3.8-max upstream fixes), #29218 (OpenAI Daybreak alias pricing), #29226 (Copilot utility models at zero), #29585 (DeepSeek V4.1 Flash, Claude Mythos 5.1, Gemini Pro and Lite tiers, cross-provider gaps).
Commits:
feat(coderd/x/chatd): the chatd portion of feat: add GPT-6 Astra to the known models catalog, price book, and chatd #28955, applied as-is. GPT-6 Astra's function calling is Responses-only and the pinned SDK predates it, soUsesResponsesAPIdefaults it to Responses, and reasoning effortnone/minimalclamp tolow. Without this the catalog would suggest a model that fails in chat on 2.37.chore:curation.jsonandoverrides.jqbrought to their exact main state, thenprices.jsonandknownModelsGenerated.jsonregenerated on this branch withmake gen/aibridge-prices. The catalog stays at its 2.37 path underChatModelAdminPanel/knownModels; themodules/aiModelsmove (refactor: move shared AI model UI intomodules/aiModels#29040) is not backported, which is why the generated artifacts are regenerated rather than cherry-picked. The Anthropic thinking-mode test pin gains the same three entries as on main.docs(coderd/x/chatd): restoresARCHITECTURE.mdto its 2.37 state; the TODO placeholders from feat: add GPT-6 Astra to the known models catalog, price book, and chatd #28955 were later replaced on main and do not belong in a backport.fix: omitsgpt-6-astrafrom theazurecuration and regenerated catalog entry. The 2.37 fantasy pin predates Astra and the Azure provider has no hook to force Responses, so an Azure Astra config would speak Chat Completions while Astra's function calling is Responses-only. main only added the Azure entry (chore: add DeepSeek V4.1 Flash and other missed models to the known models catalog #29585) after bumping fantasy (feat: add openai_config.reasoning_model override and bump fantasy #29230), which this backport does not carry.TestCuratedAstraUsesResponsesguards the curation against reintroducing it while the pin is old.Parity check against main: overrides identical; curation differs only by the omitted Azure Astra entry; the regenerated catalog and price book otherwise differ from main only by live models.dev drift on one OpenRouter DeepSeek price (models.dev is fetched at generation time, same as the weekly refresh).
Validation on this branch:
go test ./scripts/aibridgepricesgen/... ./coderd/aibridge/prices/... ./coderd/x/chatd/chatopenai/... ./coderd/x/chatd/chatprovider/...andpnpm test src/pages/AgentsPage/components/ChatModelAdminPanel/knownModels/(52 tests) pass; pre-commit hook passed on every commit.Review record: Codex review on
7939e2araised one P1 (Azure Astra transport). Confirmed by execution (TransportFor(azure, gpt-6-astra)resolves to Chat Completions on the 2.37 pin; the guard test is red on the old curation and green after) and fixed in commit 4. Codex review onfd0e070raised two P1s asking to re-add the ARCHITECTURE.md TODO placeholders; declined per the author's direction (main already carries the human-written prose at ARCHITECTURE.md L910 and L916), replied and resolved. Re-review onfd0e070: Codex OK ("Didn't find any major issues", reviewed commitfd0e070427), zero unresolved threads, all 29 check runs green.