fix(examples/voice): swap deleted gpt-4o-audio-preview → gpt-audio-mini#607
Closed
drewdrewthis wants to merge 2 commits into
Closed
fix(examples/voice): swap deleted gpt-4o-audio-preview → gpt-audio-mini#607drewdrewthis wants to merge 2 commits into
drewdrewthis wants to merge 2 commits into
Conversation
Collaborator
Author
|
[grinder] READY for human review CI: green — zero failing, zero pending (all 17 checks SUCCESS or SKIPPED-by-design) Verified by:
ACs (from verification plan):
Do NOT merge — that's your call. |
drewdrewthis
added
pr-ready
and removed
grinding
github-actions
Bot
added
the
low-risk-change
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
drewdrewthis
added a commit
that referenced
this pull request
Jun 4, 2026
…p are in #607, not this branch Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Collaborator
Author
✅ Review + prove-it: READYReview: diff is exactly the model swap ( Prove-it (live, not CI — CI skips these): The Caveat (non-blocking): 4 stale |
drewdrewthis
added a commit
that referenced
this pull request
Jun 4, 2026
… refs after model swap Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
github-actions
Bot
added
low-risk-change
drewdrewthis
added a commit
that referenced
this pull request
Jun 5, 2026
…t-gen (#610) * docs(voice/#606): expand STT/TTS doc comments and relax audio-to-text judge criteria Adds deliberate-choice rationale comments to OPENAI_STT_MODEL and OPENAI_TTS_MODEL in both JS (voice-models.ts) and Python (voice_models.py), noting no gpt-5-family transcription/TTS models exist on the public API as of 2026-06. Also documents the Python-only OPENAI_BOT_STT_MODEL gap in the TS file. Relaxes the multimodal-audio-to-text judge criteria from overly-specific assertions (exact voice gender, exact repeat phrasing) to behavioural checks (processed audio, coherent response, non-text format acknowledgement). Updates the stale skip comment to reflect the model swap in PR #607. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * fix(voice/#606): update feature-file contract counts to match post-#561/#604 reality Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * docs(voice/#606): add AC4/AC5 doc comments — STT lock rationale + TTS callable-swap pattern - openai-realtime.ts: explain why `input.transcription.model` is locked to OPENAI_STT_MODEL and not exposed as a constructor option (Realtime API only accepts transcription-class models; callers who need a different model subclass the adapter) - openai-tts.ts: document that the TTS model is not a parameter by design — the pattern is to swap the whole TTSCallable rather than parameterise this one; link to OPENAI_TTS_MODEL for the current-gen rationale Closes #606 (AC4 + AC5) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * docs(examples/voice/#606): correct stale comment — model swap + unskip are in #607, not this branch Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
The voice-to-voice example helper and the audio-to-text example pinned
`gpt-4o-audio-preview`, which OpenAI has removed (404 model_not_found
since 2026-05-19). Any user running the canonical voice example hit an
immediate 404.
Switch to `gpt-audio-mini` — OpenAI's current cost-efficient GA
audio-chat model — matching the Python twin, which already migrated
(python/scenario/config/voice_models.py:44 OPENAI_AUDIO_CHAT_MODEL,
python/examples/test_audio_to_text.py:157). Verified live: gpt-audio-mini
accepts the identical chat.completions shape (modalities:["text","audio"],
audio:{voice,format}) and returns audio. Re-ran the voice-to-voice e2e
against prod LangWatch — success: true, real 2-turn conversation, traces
landed (project_bZspxwkhCD4POvqmIgOr2).
SDK core was unaffected (OpenAIRealtimeAgentAdapter uses gpt-realtime-mini).
This closes a py↔ts example-parity gap left by #561.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… refs after model swap Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
drewdrewthis
force-pushed
the
fix/voice-example-dead-audio-model
branch
from
3593191 to
37c0343
Compare
github-actions
Bot
removed
the
low-risk-change
Contributor
|
Automated low-risk assessment This PR was evaluated against the repository's Low-Risk Pull Requests procedure and does not qualify as low risk.
This PR requires a manual review before merging. |
drewdrewthis
added a commit
that referenced
this pull request
Jun 5, 2026
… refs after model swap Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
drewdrewthis
added a commit
that referenced
this pull request
Jun 5, 2026
…e, keep+migrate supported audio examples Cohesive retirement of the legacy gpt-4o-audio-preview voice/audio example surface, folding in the model swap from #607 (cherry-picked) and superseding the unskip plan in #486. Genuinely-dead (retired): - Tombstone docs/docs/pages/examples/multimodal/voice-to-voice.mdx and testing-voice-agents.mdx -> pointer to /voice/getting-started (URLs still 200; the langwatch vocs fork has no redirect layer, so a tombstone is how we avoid 404s on previously-public URLs). - Delete the now-unused LegacyVoiceDeprecation.mdx snippet (no importers left). - (test_voice_to_voice_conversation.py already deleted in an earlier #486 commit.) Supported (kept + migrated to gpt-audio-mini): - audio-to-text.mdx / audio-to-audio.mdx kept and updated: prereq prose now names gpt-audio-mini; LegacyVoiceDeprecation banner removed (these document the CURRENT supported single-call pattern, not a legacy one). - Python test_audio_to_text.py / test_audio_to_audio.py: skip COMMENTS rewritten to the real reason (live E2E -- real OpenAI gpt-audio-mini + LangWatch backend, cost, non-deterministic audio); skipif(CI) markers retained by design. No model literal change (they route through the helper's gpt-audio-mini default). - _generated example partials regenerated to match the migrated test sources. overview.mdx voice-agents link repointed to /voice/getting-started. Why the audio tests stay CI-skipped: they are live end-to-end tests; #486's "unskip to restore CI coverage" premise was never achievable (cost + non-determinism). The right end-state is migrated-and-intentionally-skipped. Docs build: pnpm build exits 0, no broken-link/missing-import errors; tombstone routes render the /voice/getting-started pointer. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Collaborator
Author
|
Superseded by #612. The two model-swap commits from this PR ( |
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
1 participant
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.
Problem
The voice/audio examples pinned
gpt-4o-audio-preview, which OpenAI has deleted —404 model_not_foundsince 2026-05-19. Any user running the canonical voice-to-voice example (or audio-to-text) hits an immediate 404. The SDK core is unaffected (OpenAIRealtimeAgentAdapterusesgpt-realtime-mini).Fix
Swap to
gpt-audio-mini— OpenAI's current cost-efficient GA audio-chat model — matching the Python twin, which already migrated (python/scenario/config/voice_models.py:44OPENAI_AUDIO_CHAT_MODEL,python/examples/test_audio_to_text.py:157). Closes a py↔ts example-parity gap left by #561.Files changed (4 lines):
examples/vitest/tests/helpers/openai-voice-agent.ts— model literal + 2 doc-commentsexamples/vitest/tests/multimodal-audio-to-text.test.ts— model literalVerification (live, against prod LangWatch)
gpt-audio-miniaccepts the identicalchat.completionsshape (modalities:["text","audio"],audio:{voice,format}) and returns audio — confirmed via direct/v1/chat/completionscall.multimodal-voice-to-voice-conversation.test.ts: ✅success: true, real 2-turn conversation, judge passed, traces landed in prod (project_bZspxwkhCD4POvqmIgOr2).multimodal-audio-to-audio.test.ts: ✅ passes.multimodal-audio-to-text.test.ts: no longer 404s, butresult.success=falseon brittle judge criteria ("guesses it's a male voice", "says what format the input was") —gpt-audio-minidoesn't reliably volunteer all three. This is a pre-existing brittle-criteria sensitivity, not a regression from this fix; tracked in Voice STT/TTS defaults still use gpt-4o-* models — decide modernization path #606. Reverting would restore the 404 (strictly worse), so this PR keeps the swap.Draft
Left as draft — model-default modernization (the remaining
gpt-4o-*STT/TTS) is tracked separately in #606. Ready for your review/merge call.🤖 Generated with Claude Code