cohere-transcribe-arabic-07-2026
by CohereLabs
Dialect-aware Arabic ASR with Arabic-English code-switching
CohereLabs/cohere-transcribe-arabic-07-2026mixpeek://transcription@v1/cohere_transcribe_arabic_v1Overview
Cohere Transcribe Arabic is a 2B-parameter speech recognition model from Cohere Labs (July 2026) built specifically for Arabic — including regional dialects and Arabic-English code-switching, the two places general-purpose ASR models degrade hardest. On the Open Universal Arabic ASR Leaderboard it averages 25.87% WER across dialect-heavy test sets, with 5.82% WER on Common Voice Arabic.
On Mixpeek, it fills the Arabic gap in transcription pipelines: Arabic broadcast media, Gulf and Levantine dialect recordings, and mixed Arabic-English business audio become searchable text indexed alongside embeddings and faces. Pair it with voice-activity detection at ingest — the model transcribes non-speech sounds without it and does not emit timestamps or speaker labels on its own.
Architecture
Conformer-based encoder-decoder: a large Conformer encoder for acoustic representations with a lightweight Transformer decoder for token generation. Audio resampled to 16kHz. No built-in language detection, timestamps, or diarization. Apache 2.0 license.
Mixpeek SDK Integration
import { Mixpeek } from "mixpeek";
const mx = new Mixpeek({ apiKey: "API_KEY" });
// Managed: create a collection over a bucket; Mixpeek runs this model's extractor
const collection = await mx.collections.create({
namespace_id: "my-namespace",
collection_name: "my-collection",
source: { type: "bucket", bucket_ids: ["bkt_your_bucket"] },
feature_extractor: {
feature_extractor_name: "transcription",
version: "v1",
parameters: { model_id: "CohereLabs/cohere-transcribe-arabic-07-2026" },
},
});Capabilities
- Arabic dialect coverage (Gulf, Levantine, Egyptian, Maghrebi test sets)
- Arabic-English code-switching
- 5.82% WER on Common Voice Arabic; 15.54% WER on MGB-2 broadcast
- 25.87% average WER on the Open Universal Arabic ASR Leaderboard
- Compact 2B parameters under Apache 2.0
Use Cases on Mixpeek
Benchmarks
| Dataset | Metric | Score | Source |
|---|---|---|---|
| Open Universal Arabic ASR Leaderboard (avg) | WER | 25.87% | Cohere Labs, 2026 — Model Card |
| Common Voice (Arabic) | WER | 5.82% | Cohere Labs, 2026 — Model Card |
| MGB-2 (broadcast) | WER | 15.54% | Cohere Labs, 2026 — Model Card |
| MASC (clean) | WER | 19.60% | Cohere Labs, 2026 — Model Card |
Explore on Mixpeek
Compare alternatives in this category
Hand-picked tools & platforms compared
Deep-dive technical guide
See how Mixpeek runs models as extractors
Store & search embeddings at scale
Usage-based pricing for pipelines
Compare models, APIs & infrastructure
Specification
Research Paper
Cohere Transcribe Arabic Model Card
arxiv.orgBuild a pipeline with cohere-transcribe-arabic-07-2026
Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.
Run on your data, free