Repository navigation
fix(ai): sync pi e5fe9777 and cedefc53 - #3829
Merged
Merged
Conversation
…5fe9777) models.dev removed the retired Copilot models (claude-opus-4.6/4.7, gpt-5.4, gpt-5.5) and the Together reasoning-only models (DeepSeek-R1, MiniMax-M2.7). The model generator no longer carries special cases for them: the Copilot extended-context list and the claude-opus-4.7 thinking-level override lose the retired IDs, and the Together reasoning-only compat and fixed thinking-level map are gone. Generated catalogs stay uncommitted. Most of upstream's change already landed in Atomic and is not repeated: - Copilot default gpt-6.1-sol and its changelog entry: 8565667 (#3819), in model-resolver-defaults.ts where Atomic keeps the defaults. - Baseten Kimi-K2.6 test and image-tool-result block removal: 2e217d5 (#3816). - Together MiniMax-M2.7 check removal: 48f6fe2 (#3815). - Copilot opus-4.7 assertion removal, and the gpt-6-luna swaps in openai-responses-tool-result-images: 473ed00 and 8565667. Ported here: the generator cleanup; the Copilot stream e2e block now targets gpt-6-luna via OpenAI Responses and reuses it for the thinking tests; the tool-call-id normalization comment names gpt-6.1-sol, the model Atomic's test uses. Atomic's openai-responses-compat test already uses gpt-6-luna (also off: "none"), so only upstream's #2567 comment is added, naming gpt-6-luna. Upstream: earendil-works/pi@e5fe977 Assistant-model: Claude Opus 5.5 Assistant-workflow: issues-to-merged-prs (run bc9989bb-3139-46c4-89d5-cd60cb951cfe) Assistant-verification: generate-models passed: node scripts/generate-models.ts --strict --json-only to /tmp completes; github-copilot output has no claude-opus-4.6/4.7, gpt-5.4 or gpt-5.5 and together no MiniMax-M2.7 or DeepSeek-R1 Assistant-verification: vitest focused passed: packages/ai stream, openai-responses-compat, tool-call-id-normalization, github-copilot-anthropic, together-models, baseten-models, openai-responses-tool-result-images, image-tool-result (4 files ran, 69 tests; 4 credential-gated files skipped) Assistant-verification: typecheck passed: packages/ai npm run typecheck (448 files) Co-authored-by: Norin Lavaee <nlavaee@umich.edu>
…i cedefc53) On a ChatGPT Business plan the workspace admin sets a per-user usage limit, and the Codex backend reports it as subscription_sharing_workplace_individual_usage_limit_exceeded. The code was treated as a transient error, and the message gave no way forward. It is now non-retryable in isRetryableAssistantError, and the OpenAI Responses error message tells the user to ask their workspace admin to raise the limit at admin.openai.com. Adaptation: the changelog entry goes under Atomic's existing Unreleased "Fixed" subsection (upstream used "Added"), and its wording says the error is no longer classified as retryable. Request-level HTTP 429 retries in provider-retry.ts are unchanged, as upstream. Tests: upstream's retry.test.ts case is ported. Upstream had no test for the message, so test/openai-responses-compat.test.ts gains a mocked-429 case for both ChatGPT usage-limit codes; the existing subscription_sharing_usage_limit_exceeded hint had no coverage either. Upstream: earendil-works/pi@cedefc5 Assistant-model: Claude Opus 5.5 Assistant-workflow: issues-to-merged-prs (run bc9989bb-3139-46c4-89d5-cd60cb951cfe) Assistant-verification: vitest red-green passed: with src/utils/retry.ts and src/api/openai-responses.ts stashed, retry.test.ts (expected true to be false) and the workspace-limit message case fail (2 failed, 100 passed); with the fix 102/102 pass Assistant-verification: vitest packages/ai full suite passed: 192 files passed, 27 credential-gated skipped, 1997 tests passed, 863 skipped Assistant-verification: npm run check passed: biome (8 infos in untouched files, no errors), bun typecheck for root and coding-agent, shrinkwrap check Assistant-duration: 7m converged, estimated 45m Co-authored-by: Norin Lavaee <nlavaee@umich.edu>
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
Moves the earendil-works/pi#2567 reference from a comment into the test name so it appears in test results, as AGENTS.md requires. Assistant-model: Claude Opus 5.5 Assistant-workflow: issues-to-merged-prs (run bc9989bb-3139-46c4-89d5-cd60cb951cfe) Assistant-verification: vitest focused passed: packages/ai test/openai-responses-compat.test.ts (60 tests) Assistant-verification: npm run check passed: biome, bun typecheck passes, shrinkwrap check Co-authored-by: Norin Lavaee <nlavaee@umich.edu>
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
Syncs Atomic with two upstream pi commits, one Atomic commit each.
generate-models.tscleanup (Together reasoning-only maps, retired Copilot IDs),stream.test.tsCopilot block ongpt-6-luna, stale comments. Skipped (already in Atomic): Copilot defaultgpt-6.1-sol(8565667, #3819), Baseten Kimi-K2.6 tests (2e217d5, #3816), Together MiniMax-M2.7 check (48f6fe2, #3815), Copilot opus-4.7 assertions (473ed00), tool-result-images model swap. No changelog: nothing user-facing.subscription_sharing_workplace_individual_usage_limit_exceededis non-retryable; the error message tells the user to ask their workspace admin to raise the limit. Changelog entry goes under existing Unreleased### Fixed. Added a message test (upstream had none).Verification
retry.tsandopenai-responses.tsreverted,retry.test.tsand the new message test fail (2 failed, 100 passed); with the fix 102/102 pass.packages/aifull vitest: 1997 passed, 863 credential-gated skipped.npm run check: passed.--strict --json-only) completes; retired model IDs absent.provider-retry.tsis unchanged, as upstream.Assistant-workflow: issues-to-merged-prs (run bc9989bb-3139-46c4-89d5-cd60cb951cfe)
Assistant-verification: vitest red-green passed: retry and message tests fail before, pass after
Assistant-verification: npm run check passed
Assistant-verification: GitHub Actions CI passed: test (all platforms) on 49c5765
Assistant-verification: Greptile review passed: confidence 5/5, no new issues on 49c5765
Assistant-duration: 36m converged, estimated 2h
Co-authored-by: Norin Lavaee nlavaee@umich.edu
User-preference: loop each PR until CI is green and Greptile has no feedback, then merge
The changes since the previous review only update a test name and remove its redundant comment; no outstanding findings block merging.Summary
Syncs upstream model cleanup and ChatGPT Business usage-limit support.
No new issues found in the changes since the previous review.
Reviews (2) · Last reviewed commit: "test(ai): name upstream issue in Copilot..." · Reviewed by Greptile