Cohere Launches Open Weights Model Command A More Than A Year Since The Command A Release
Source
artificialanalysis.ai
Author
Artificial Analysis articles
Date
Why it matters
Command A+ ranks first for non-hallucination at 86% and runs near 281 tokens per second, though it trails on hard coding and science.
Key takeaways · AI-distilled
Command A+ scores 37 on the Intelligence Index, in line with Claude 4.5 Haiku. Its 86% non-hallucinationWhen a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.Full definition → rate leads by about 3 points, but 9% accuracy leaves its AA-Omniscience score at -4, a model that knows its limits.
On Cohere's API it runs at about 281 output tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → per second, faster than GPT-5.4 nano, Claude 4.5 Haiku and Grok 4.3, but slower than Gemini 3.1 Flash-Lite Preview at 304.
It trails peers on hard science (HLE about 11%, GPQA Diamond about 76%) and coding (Terminal-Bench Hard about 25%, SciCode about 38%), while scoring 63% on MMMU-Pro visual reasoning, between Claude 4.5 Haiku and GPT-5.4 nano.
Terms in this piece · Glossary
hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
