Cohere launches Command A+, its 'most powerful' LLM optimized for minimal hardware

Cohere announced Command A+ on May 20, with Hugging Face's official account retweeting the launch — a notable amplification given Cohere's enterprise-Canadian positioning historically getting less retail-developer attention than DeepSeek, Qwen, or Mistral releases. Cohere's pitch in the announcement copy is two-pronged: it is the company's most powerful model to date and is explicitly engineered to run on minimal hardware footprints, targeting enterprises that need on-prem or VPC deployment.

The 'minimal hardware' framing is competitively pointed. DeepSeek V4 Pro shipped this week at 1.6T total parameters (depth play), and Alibaba's Qwen3.7-Max-Preview ranked top-13 globally on LM Arena text. Cohere is going the other direction: high quality at a hardware envelope an enterprise can actually deploy without standing up a frontier-scale cluster. That's the same niche Mistral has historically owned in Europe.

The ecosystem signal worth noting: Cohere CEO Aidan Gomez's social posture this week ('Canada is the greatest country on earth') paired with the Command A+ launch reads as a deliberate sovereign-AI positioning play, similar to Mistral's European bank pitch this week. With OpenAI ending Microsoft exclusivity and Anthropic acquiring developer-tooling layers like Stainless, the 'neutral enterprise alternative' positioning is suddenly contested space — and Cohere is reasserting it.

What to watch: published benchmarks (Cohere's announcement was high-level — no leaderboard numbers in the social blurb), pricing per million tokens for the hosted API, and whether the weights will be released openly on Hugging Face or held as API-only. Given Hugging Face's amplification, an open-weight or open-weight-research release seems likely, which would significantly change Command A+'s reach versus a closed-API launch.