vllm, the popular open-source LLM inference engine, released v0.22.0 on May 29, 2026.

vllm, the popular open-source LLM inference engine, released v0.22.0 on May 29, 2026. The release notes specifically include two enhancements for Cohere models: 'enable Cohere MoE (#43143)' and 'pipeline parallelism for Cohere vision (#42819)'. These additions mean that Cohere's Mixture-of-Experts and vision models can now be efficiently served using vllm, improving inference speed and scalability. This integration is significant because vllm is widely adopted in the AI infrastructure ecosystem, and explicit support for Cohere's architecture signals growing community and developer interest. It also reduces friction for enterprises deploying Cohere models in production, potentially increasing demand for Cohere's platform. While this is a technical update rather than a direct business announcement, it reflects positive momentum for Cohere's ecosystem adoption. Source: vllm GitHub Releases