Uploaded September 2025 | Updated September 2026, 29 minutes ago
Cohere just pulled a wild move: a 111B parameter reasoning model that runs on a single GPU and beats GPT-4o on throughput. ⚡️
It’s called Command A Reasoning (open weight, research license). Designed for enterprises, it supports 23 languages, tool use, agent workflows, and comes with a configurable reasoning mode. The kicker? You can run it on one H100/A100 GPU with 128k context, or scale to 256k on two GPUs. That’s insane efficiency for a model this size.
Performance? Around 156 tokens/sec, nearly 2x faster than GPT-4o. It also crushed tau-bench (a benchmark for tool-agent-user interaction in real-world domains).
The catch: full commercial use requires a license via Cohere’s sales team—but it’s free for research. This makes it attractive for Canadian companies seeking a local LLM alternative to US giants, while still being a big win for the research community.
Cool to see Cohere staying competitive at the top of the LLM game. What do you think—game changer or research flex?
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#Cohere #AInews #LLM #short
Cohere just pulled a wild move: a 111B parameter reasoning model that runs on a single GPU and beats GPT-4o on throughput. ⚡️
It’s called Command A Reasoning (open weight, research license). Designed for enterprises, it supports 23 languages, tool use, agent workflows, and comes with a configurable reasoning mode. The kicker? You can run it on one H100/A100 GPU with 128k context, or scale to 256k on two GPUs. That’s insane efficiency for a model this size.
Performance? Around 156 tokens/sec, nearly 2x faster than GPT-4o. It also crushed tau-bench (a benchmark for tool-agent-user interaction in real-world domains).
The catch: full commercial use requires a license via Cohere’s sales team—but it’s free for research. This makes it attractive for Canadian companies seeking a local LLM alternative to US giants, while still being a big win for the research community.
Cool to see Cohere staying competitive at the top of the LLM game. What do you think—game changer or research flex?
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#Cohere #AInews #LLM #short
