cohere-transcribe-03-2026

by CohereLabs

#1 on Open ASR Leaderboard with 14-language support

286Kdl/month

2Bparams

HuggingFace Use in Pipeline

Identifiers

Model ID

CohereLabs/cohere-transcribe-03-2026

Feature URI

mixpeek://transcription@v1/cohere_transcribe_03_v1

Overview

Cohere Transcribe is a 2B-parameter automatic speech recognition model that ranks #1 on the Open ASR Leaderboard for English. Trained on 500K hours of audio data, it delivers 3x faster real-time processing compared to models of similar accuracy. The model supports 14 languages with strong multilingual performance.

For multimodal search pipelines, accurate transcription is foundational -- every word in the transcript becomes searchable text. Higher transcription accuracy directly translates to better full-text search over audio and video content.

Architecture

Encoder-decoder architecture optimized for streaming and batch ASR. 2B parameters trained on 500K hours of diverse audio. Supports NeMo framework for enterprise deployment.

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

await mx.collections.ingest({
  collection_id: "my-collection",
  source: { url: "https://example.com/podcast.mp3" },
  feature_extractors: [{
    feature: "transcription",
    model: "CohereLabs/cohere-transcribe-03-2026"
  }]
});