[bot] Cohere: Audio Transcription API (`client.audio.transcriptions.create()`) not instrumented

[braintrust-bot]

Summary

The Cohere Audio Transcription API (client.audio.transcriptions.create()) is not instrumented. Calls to this endpoint produce zero Braintrust tracing. This is a speech-to-text inference endpoint that runs the cohere-transcribe-03-2026 ASR model (2B parameter Conformer architecture) on audio files and returns transcribed text.

The Braintrust Cohere integration currently instruments chat, chat_stream, embed, and rerank (V1 and V2), but has no handling for the audio transcription surface introduced in Cohere SDK v6.1.0.

What is missing

Cohere Resource	Method	Instrumented?
client.chat() / client.v2.chat()	sync + async	Yes
client.chat_stream() / client.v2.chat_stream()	sync + async	Yes
client.embed() / client.v2.embed()	sync + async	Yes
client.rerank() / client.v2.rerank()	sync + async	Yes
client.audio.transcriptions.create()	sync	No
client.audio.transcriptions.create() (async)	async	No

API details

• Endpoint: POST /v2/audio/transcriptions
• Model: cohere-transcribe-03-2026
• Supported formats: FLAC, MP3, MPEG, MPGA, OGG, WAV (25MB max)
• Languages: 14 languages (English, German, French, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Arabic, Vietnamese, Chinese, Japanese, Korean)
• Parameters: model, language (ISO-639-1), file, temperature
• Return type: AudioTranscriptionsCreateResponse
• SDK: Both TranscriptionsClient.create() and AsyncTranscriptionsClient.create() exist in the cohere package at src/cohere/audio/transcriptions/client.py

Minimum instrumentation

At minimum, both sync and async create() methods should create spans capturing:

• Input: file metadata (name, size, format), language, model
• Output: transcribed text
• Metadata: provider: "cohere", model, language, temperature
• Metrics: latency

Precedent in this repo

Several other integrations instrument equivalent audio transcription endpoints:

• OpenAI: AudioTranscriptionsPatcher in py/src/braintrust/integrations/openai/patchers.py
• Mistral: TranscriptionsPatcher in py/src/braintrust/integrations/mistral/patchers.py
• LiteLLM: LiteLLMTranscriptionPatcher in py/src/braintrust/integrations/litellm/patchers.py

Braintrust docs status

not_found — Cohere is not listed on the Braintrust integrations page. There is no dedicated Cohere integration docs page.

Upstream sources

• Cohere Transcribe documentation: https://docs.cohere.com/docs/transcribe
• Cohere Python SDK on PyPI (v6.1.0): https://pypi.org/project/cohere/
• Cohere Python SDK source: https://github.com/cohere-ai/cohere-python (src/cohere/audio/transcriptions/client.py)
• Feature released approximately March 2026; described as an "open source research release"

Local files inspected

• py/src/braintrust/integrations/cohere/patchers.py — defines patchers for Chat, ChatStream, Embed, Rerank (V1 and V2 variants); zero references to audio, transcription, or transcribe
• py/src/braintrust/integrations/cohere/tracing.py — wrapper functions for chat, embed, rerank; no audio/transcription wrappers
• py/src/braintrust/integrations/cohere/integration.py — integration class registers chat/embed/rerank patchers; no TranscriptionPatcher
• py/src/braintrust/integrations/cohere/test_cohere.py — no audio/transcription test cases
• py/pyproject.toml — Cohere version matrix: latest pinned to cohere==6.1.0, older pin at 5.0.0