Command A+
Cohere's first MoE (218B/25B active), Apache 2.0, 48 languages including every EU one, runs on two H100s. Built for sovereign enterprise deployments with citation grounding, not for topping leaderboards. Hosted pricing is contact-sales.
Key facts
- Score
- 33 / 100, rank #25Evals22Pricen/aPeoplen/aFit70
- Price
- No verified list price
- Context window
- 128K
- Open weights
- Yes
- Released
- May 20, 2026
- Best for
- Sovereign/on-prem enterprise RAG, EU multilingual translation and support, Citation-grounded search
Cohere has not published a list price for Command A+ that we could verify from a primary source.
Independent evals
| Benchmark | Result | Who ran it | Read |
|---|---|---|---|
| Artificial Analysis Intelligence Index | 13 | Command A+ - Artificial Analysis | Oct 10, 2026 |
- Artificial Analysis output speed (tokens/s): 229.4 (Command A+ - Artificial Analysis)
Caveats
- AA 13 on the current index; Cohere's own launch claim of 37 was on an older index version.
- 128K context is the smallest in this list.
- No public per-token price for production keys.
Questions people ask
How much does Command A+ cost?
Cohere has not published a list price for Command A+ that we could verify from a primary source.
How good is Command A+?
Command A+ ranks #25 of 26 on the Models at Work leaderboard with a score of 33 out of 100 (October 2026 edition). Cohere's first MoE (218B/25B active), Apache 2.0, 48 languages including every EU one, runs on two H100s. Built for sovereign enterprise deployments with citation grounding, not for topping leaderboards. Hosted pricing is contact-sales.
What is Command A+ best for?
Command A+ is best for sovereign/on-prem enterprise rag, eu multilingual translation and support, citation-grounded search.
What are Command A+'s benchmark scores?
Artificial Analysis Intelligence Index: 13 (Command A+ - Artificial Analysis).
What is Command A+'s context window?
Command A+ has a 128K token context window according to Cohere.
Is Command A+ open weights?
Yes. Command A+ is released with open weights, so it can be self-hosted.
Compare Command A+
Methodology: Score is 0 to 100 and computed, not typed: 55% independent evals (Artificial Analysis, Arena, Scale SEAL, Epoch, each scaled against its natural floor and the best score in this table, then averaged), 15% blended price on a fixed log scale ($0.05 per million is 100, $60 is 0), 15% practitioner sentiment from Signals and community threads, 15% operational fit (context window, open weights). A missing component drops out and the rest are reweighted. A model with no independent eval yet is provisional and ranks below every measured one. Within three points is a tie. Vendor-published figures are labelled and never counted.