This is @Cohere’s fastest and most powerful model yet. It uses a 218B MoE architecture with 25B active parameters, built for enterprise agents that need reasoning, tool use, vision, long context, and multilingual workflows.
Command A+ can run on as little as two H100 GPUs with W4A4 quantization, which is pretty impressive considering its overall performance across agentic, multimodal, and multilingual tasks.
Love how they put it:
We want to give developers direct access to enterprise-grade agentic capabilities from experimentation to production.
Open source AI workspaces feel like an important counterbalance in the ecosystem.
Report
how does Cohere perform on long-running agent workflows, not just isolated benchmarks. When an enterprise agent has large context, tool calls, retries, vision inputs, and multilingual reasoning in the same run, where do you see the biggest strength, reasoning quality, latency, cost efficiency, or reliability?
Replies
Flowtica Scribe
Hi everyone!
Apache 2.0 on Command A+ is a big deal!
This is @Cohere’s fastest and most powerful model yet. It uses a 218B MoE architecture with 25B active parameters, built for enterprise agents that need reasoning, tool use, vision, long context, and multilingual workflows.
Command A+ can run on as little as two H100 GPUs with W4A4 quantization, which is pretty impressive considering its overall performance across agentic, multimodal, and multilingual tasks.
Love how they put it:
Kudos!
Rizzle AI
Open source AI workspaces feel like an important counterbalance in the ecosystem.
how does Cohere perform on long-running agent workflows, not just isolated benchmarks. When an enterprise agent has large context, tool calls, retries, vision inputs, and multilingual reasoning in the same run, where do you see the biggest strength, reasoning quality, latency, cost efficiency, or reliability?