Cohere Slashes Command-R+ Pricing by 60% — and the Race to the Bottom Gets Faster
Cohere cut Command-R+ to $0.003 per 1K input tokens — matching Claude Sonnet and undercutting GPT-4 Turbo by 70%, adding more heat to an already intense pricing war.
Cohere just made enterprise AI a lot cheaper. The company slashed pricing on its Command-R+ model by roughly 60%, dropping input token costs to $0.003 per 1K tokens — putting it on equal footing with Anthropic’s Claude Sonnet and well below GPT-4 Turbo, which still runs at $0.01 per 1K input tokens. For anyone running high-volume enterprise workloads, that gap is no longer theoretical savings — it’s a budget line item.
The move lands at a moment when the AI pricing landscape is changing faster than most finance teams can update their spreadsheets. Open-source models keep narrowing the gap with proprietary ones, and every quarter brings another round of cuts from someone trying to gain ground. Cohere isn’t just responding to the market — it’s trying to reshape it.
The Numbers That Actually Matter
At $0.003 per 1K input tokens, Command-R+ now matches Claude Sonnet’s input pricing head-on. Anthropic charges $0.015 per 1K output tokens for Sonnet — Cohere’s output pricing on Command-R+ sits competitively below that. GPT-4 Turbo from OpenAI is in a different bracket altogether at $0.01 input and $0.03 output per 1K tokens. If you’re building a retrieval-augmented generation (RAG) system at scale — which is exactly the workload Command-R+ is designed for — these numbers compound fast. A system processing 100 million tokens a month saves thousands of dollars monthly by switching from GPT-4 Turbo to Command-R+ at these rates.
Cohere has always positioned Command-R+ as an enterprise model built specifically for RAG and tool use rather than general-purpose chat. The price cut doesn’t change what the model does — it changes who can afford to do it at scale.
Why Now, and Why It Stings
The timing isn’t random. Open-source alternatives like Meta’s Llama series and Mistral’s models have gotten capable enough that enterprise buyers are now using them as a credible negotiating chip — or actually deploying them. When the free option is genuinely competitive, the paid option needs a reason to exist. Cohere’s answer is infrastructure, support, compliance features, and now, a price that removes cost as an objection.
For OpenAI and Anthropic, this is the kind of move that quietly accelerates margin compression without requiring a direct confrontation. Neither company needs to respond tomorrow. But if Cohere converts even a fraction of GPT-4 Turbo’s enterprise volume at these rates, the downstream effect on Q2 2026 revenue projections gets uncomfortable. Google’s Gemini Flash already plays aggressively at the low end of the market — now there’s pressure from the enterprise-specialist tier too.
Mistral has been competitive on price for a while, so this move is less about undercutting them and more about Cohere closing the gap and making the comparison chart harder to ignore. When two models are priced the same, buyers start reading the fine print on context windows, latency, and RAG performance — and that’s a fight Cohere has been preparing for.
What This Means for the Market
The enterprise AI pricing floor keeps dropping, and it’s not slowing down. Every major lab is now caught between two forces: users who want frontier-model performance and buyers who want commodity-model prices. Cohere is betting that specialization — being the best at a specific, high-value use case rather than trying to be everything — is worth more than matching GPT-5’s benchmark scores.
Whether that bet pays off depends on how many enterprise procurement teams are willing to move off incumbents for a 60% cost reduction. History suggests a lot of them will at least run the pilot. And if Command-R+ performs on par in production RAG environments — which Cohere’s existing customer base suggests it does — the switching cost starts looking pretty manageable against those savings.
OpenAI and Anthropic haven’t announced any pricing responses as of late February 2026. They don’t have to react immediately. But the companies that ignore sustained price pressure long enough tend to find the ground has shifted when they finally look down.
