Command A+
Cohere·
LLMsopen-weightcloud + localMulti-GPU Self-Host
Overview
218B sparse Mixture-of-Experts (~25B active) open-weight flagship under Apache 2.0, runnable on as few as 2x H100 GPUs. Enterprise- and RAG-focused with strong multilingual coverage. Available as open weights and via Cohere's API. Early independent benchmarks (Artificial Analysis): GPQA Diamond ~76, HLE ~11, MMMU-Pro 63, Intelligence Index ~37 — flagged as estimates, AA's full evaluation is still ongoing.
Capabilities and innovations
218B sparse MoE / ~25B activeApache 2.0 open weightsMultilingualEnterprise / RAG focusRuns on 2x H100Apache 2.0 open-weight enterprise flagship
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 76% | third party GPQA Diamond ~76 — early Artificial Analysis estimate; AA's full independent evaluation is still listed as forthcoming (Intelligence Index shown 29-37 across AA surfaces). |
| Humanity’s Last Exam (no tools) | 11% | third party HLE ~11 — early Artificial Analysis estimate (no-tools). |
| MMMU-Pro | 63% | third party MMMU-Pro 63 — Artificial Analysis. |
- SWE-bench Pro: independent evaluation pending — No comparable coding score published for Command A+; Cohere's coding-focused model is North Mini Code.
- Terminal-Bench 2.x: independent evaluation pending
- MMLU-Pro: not reported by the vendor
Architecture and hardware
- Parameters
- 218B total, 25B active per token (MoE)
- Estimated VRAM at Q4
- ~126 GB, Multi-GPU Self-Host class
Reliability
- Hallucination rate (Vectara HHEM)
- 14% (lower is better)independent
Links
More from Cohere
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.