Cohere: AI models, benchmarks & research

Cohere is an AI research organization that develops large language models and other AI systems. They have released 7 models across 0 model families.

Cohere maintains GPQA, Global-MMLU-Lite, Global-MMLU, Cohere Agentic Question Answering, Cohere Data Analysis, Cohere Memory Usage Quality and 5 more. The Benchmarks tab shows each one with its rating (whether it counts toward the LLM Stats Score, a category index, or is listed only) and the number of models scored.

Cohere has released 7 AI models tracked on LLM Stats. See the Models tab for the full catalog with technical specifications, benchmark scores, and pricing.

Yes. 4 of Cohere's 7 models are released under open-weight licenses, while 3 are proprietary. Check the Models tab for per-model license details.

Cohere's models have been evaluated on 43 benchmarks. Their top-performing model is North Micro Vision Instruct on DocVQA with a score of 0.9. See the Benchmarks tab for the full breakdown.

Cohere's most affordable model is Embed 5 Fast with input pricing from $0.08 per 1M tokens via commercial API providers. 2 of their models have public pricing. See the Pricing tab for the full comparison.

Cohere has been releasing AI models since August 2024. Their latest release was in September 2026. See the Timeline tab for the full release history.

Visit https://cohere.ai for official documentation and API access. Many of Cohere's models are also available through third-party API providers — check the Pricing tab to compare pricing across providers.