Cohere has developed an open source text format specifically for writing

Enterprise AI company Cohere on Thursday launched its first speech model: Write is an open-source automatic speech recognition system that can be used for text writing and speech analysis.

Slightly lighter at just 2 billion units, this model is designed to be used with consumer grade GPUs for those who want to own one. It currently supports 14 languages: English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Chinese, Japanese, Korean, Vietnamese, and Arabic.

Cohere says Transcribe beats models like Zoom Scribe v1, IBM Granite 4.0 1B, ElevenLabs Scribe v2, and Qwen3-ASR-1.7B Speech on the leading Hugging Face Open ASR teamachieving an average word error rate (WER) of 5.42, lower than any other model on the benchmark.

The company claims that Transcribe had a 61% higher success rate than other brands when reviewers rated their transcripts as accurate, consistent, and useful. However, the model fell behind its rivals when it had to write Portuguese, German, and Spanish.

Cohere says Transcribe can process up to 525 minutes of text per minute, which is the highest in its model range.

The company plans to integrate Transcribe into its calling platform, Northand it makes the brand available through its own API for free. The model will also be available at Model Vaultplatform powered by Cohere.

Speech recognition models are becoming increasingly popular as the demand for text-to-speech software such as Granola and Whisper Flow.

Earlier this year, Cohere says he said which investors say is generating $240 million in revenue by 2025, and its CEO, Aidan Gomez, is said to be starting. may be publicly available “soon”.