Command A+
AvailableCohere's open-source enterprise model, released May 20 2026 (API name command-a-plus-05-2026) under Apache 2.0, and described by Cohere as the last model in the Command A family. Its first mixture-of-experts model: 218B total and 25B active parameters. Text and image input with text, reasoning and tool-use output; 128K input context and 64K max generation; 48 languages including all official EU languages. Unifies vision, reasoning, translation and agentic capabilities of earlier Command A models; Cohere reports up to 110% higher throughput and 30% lower latency than Command A Reasoning, and tau2-Bench Telecom up from 37% to 85% (vendor-reported). Weights on Hugging Face in BF16, FP8 and W4A4; runs on 2x H100 or 1x B200 at W4A4. Also on the Cohere API and Model Vault.
Specifications
- License
- Open source · Apache-2.0
- Weights
- Downloadable
- Architecture
- Mixture-of-Experts
- Parameters
- 218B · 25B active
- Context window
- — tokens
- Max output
- —
- Knowledge cutoff
- —
- Price (in / out, $/M)
- —
- Modalities
- TextVision
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.