release: 1.8.0

[openai-sdks]

Automated Release PR

1.8.0 (2026-08-13)

Features

• api: Add gpt-daybreak-blue-latest, gpt-daybreak-red-latest, and (ed0e180)
• api: add new Responses model identifiers (#51) (ed0e180)
• api: add workload identity access token issued event (#55) (75ff54a)

Bug Fixes

• api: clarify audio upload metadata requirements (#52) (17c793d)
• api: handle nullable stream flags (#53) (fa22e6e)

Chores

• api: Update generated-source attribution in file headers (1b6ec73)
• api: Update generated-source attribution in file headers (#49) (1b6ec73)
• remove Stainless attribution and infrastructure (#54) (2e59ae5)

---

This PR was generated with Release Please. See documentation.

[jbeckwith-oai]

CHANGELOG.md lists the only 1.7.2 change twice: both bullets have the same description and point to the same 1b6ec73 commit, with the second merely adding the #49 link. This duplicate will flow directly into the published GitHub release notes. Please collapse these into one entry (preferably the PR-linked form), and fix the release metadata/regeneration input if needed so Release Please does not reintroduce it.

[jbeckwith-oai]

The release-note duplication is still present on this regenerated head: both 1.7.2 bullets describe the same change and point to the same 1b6ec73 commit, with the second only adding the #49 link. Publishing this release would therefore duplicate the sole change in the GitHub release notes. Please collapse these into one PR-linked entry and fix the Release Please input/history if necessary so regeneration keeps it deduplicated. CI is green; this remains the only blocker.

[jbeckwith-oai]

The release-note blocker remains: the 1.7.2 changelog lists the sole 1b6ec73 change twice. Both bullets describe the same generated-source attribution update and point to the same commit; the second merely adds the #49 link. This will duplicate the only change in the published release notes. Please collapse these into one PR-linked bullet and fix or regenerate the Release Please source so the entry stays deduplicated. The version metadata, full v1.7.1 release delta, formatting, vet, build, diff hygiene, and file-size thresholds otherwise look clean; exact-head GitHub test, lint, and analysis checks are green (build-artifacts was still running at review time).

[jbeckwith-oai]

CHANGELOG.md still duplicates both commits in this two-commit release. Lines 8-9 point to the same ed0e180 commit twice; line 8 is also truncated after “and”, omitting gpt-5.6-cyber and the rest of the summary. Lines 14-15 likewise repeat the same 1b6ec73 attribution change, with the second bullet only adding the #49 link. Publishing this would produce four bullets for two changes, including an incomplete feature headline. Please collapse each pair to one PR-linked entry (a complete #51 feature bullet naming all three model identifiers, and one #49 attribution chore) and fix or regenerate the Release Please input/history so the duplication does not return. The full release delta, version metadata, formatting, vet, build, tests, artifact matrix, diff hygiene, and file-size thresholds otherwise look clean; all exact-head CI checks are green.

[jbeckwith-oai]

CHANGELOG.md still has the unresolved duplicate and truncated release notes. Lines 8-9 both describe and link to ed0e180; line 8 is truncated after “and” and omits gpt-5.6-cyber, while line 9 is only the generic PR-linked summary. Lines 19-20 likewise repeat 1b6ec73, with line 20 merely adding the #49 link. With the correctly single #52 entry, this produces five bullets for three source commits and would publish two duplicated changes plus an incomplete feature headline. Please collapse each pair into one PR-linked entry: a complete #51 feature bullet naming all three model identifiers, and one #49 attribution chore. Also fix or regenerate the Release Please input/history so the duplication does not return on the next head. The full v1.7.1 release delta, 1.8.0 version metadata, generated changes, formatting, vet, build, diff hygiene, and file-size thresholds otherwise look clean; exact-head lint, test, and analysis checks are green, with build-artifacts still running.

[jbeckwith-oai]

Two actionable issues remain on exact head 41ae6a351fd67d0cc068af6ee896e19cbe40ad4e:

1. The prior release-note blocker is still present. CHANGELOG.md records six bullets for four actual release commits: lines 8–9 both point to ed0e180, while lines 20–21 both point to 1b6ec73. The first ed0e180 bullet is also truncated after “and,” omitting gpt-5.6-cyber:

   openai-cli/CHANGELOG.md

   Line 8 in 41ae6a3

   * **api:** Add gpt-daybreak-blue-latest, gpt-daybreak-red-latest, and ([ed0e180](https://github.com/openai/openai-cli/commit/ed0e18007e860032896c7e8a2c3a1b6d6bdb20de))

   and

   openai-cli/CHANGELOG.md

   Line 20 in 41ae6a3

   * **api:** Update generated-source attribution in file headers ([1b6ec73](https://github.com/openai/openai-cli/commit/1b6ec733a7033caa99d2c06ac22d37c6e75d34f8))

   . Please keep one complete PR-linked entry per commit: a complete #51 feature entry naming all three models and the existing #49 chore entry. The malformed unlinked entries come from the conventional-commit-formatted Summary lines in PRs #51 and #49, so also correct the supported Release Please input/override path (and avoid that Summary shape in future source PRs) so regeneration does not restore them. Rewriting merged history is not required.

2. The nullable-stream fix changes ten handlers to recognize an explicit *bool(true), but the generated repository tests still exercise only --stream=false; for example:

   openai-cli/pkg/cmd/response_test.go

   Line 41 in 41ae6a3

   "--stream=false",

   . The old broken cmd.Bool("stream") implementation also passed those false-path tests, so the regression is not durably covered here. Please add one representative test using --stream=true with an SSE response and assert observable streaming/iteration behavior. One endpoint is sufficient; duplicating it across all ten would be unnecessary.

The four-commit release delta is otherwise coherent: version/manifest metadata agrees on 1.8.0, the generated overlay/provenance model is consistent, no dependency/security/architecture or <1k→>1k regression was found, local gofmt/vet/build/internal tests/Windows cross-compilation and stream true/false/null smoke checks passed, and exact-head lint/test/build/artifact/CodeQL CI is green.

[jbeckwith-oai]

Two prior blockers remain unresolved on this exact release head:

1. CHANGELOG.md:8-9,20-21 still contains six bullets for four source commits. ed0e180 and 1b6ec73 each appear twice, and the first ed0e180 headline is truncated after “and”, omitting gpt-5.6-cyber. Please keep one complete PR-linked entry for #51 naming all three identifiers and one entry for #49, then correct the merged PR metadata using Release Please's supported BEGIN_COMMIT_OVERRIDE / END_COMMIT_OVERRIDE input so regeneration preserves the fix.

2. The nullable-stream change in fa22e6e still changes ten branches across eight handlers without a single --stream=true regression test. All existing generated cases pass --stream=false, which also passed the old broken cmd.Bool path, so green CI never exercises the new true branch. Please add one representative --stream=true SSE test that asserts the emitted/iterated event (one endpoint is sufficient; no generated duplication needed), then regenerate this release PR.

The version bump and other release metadata are coherent, and exact-head CI is green; I found no additional issues.