Command A AI Evaluation | LayerLens - Benchmark Results & Performance

Model

Command A by Cohere

text->text

Free credits included. No credit card required.

Command A is Cohere’s advanced assistant model, released in March 2025. It is optimized for enterprise applications, excelling in tasks such as content generation, analysis, tool use, retrieval-augmented generation (RAG), and multilingual support. The model features 111 billion parameters and a 256K token context window, offering high performance with efficient hardware requirements.

Evaluations

across 26 benchmarks

Latency

724ms

Context length

256k

tokens

Cost

$2.50 · $10

input · output per 1M tokens

Key takeaways

Command A is Cohere's advanced assistant model, released in March 2025, optimized for enterprise applications (content generation, analysis, tool use, RAG, multilingual support).
It is a Transformer-based model with 111 billion parameters and a 256K token context window, featuring open weights and designed for efficient hardware use.
The model demonstrates strong performance in easy-level scientific reasoning, achieving 98.82% accuracy on the AI2 Reasoning Challenge - Easy, and adheres to output formatting instructions.
Command A struggles significantly with complex reasoning, mathematical, scientific, and symbolic manipulation tasks, scoring only 3.72% on the 'Humanity's Last Exam' benchmark.
Its limitations include difficulties with precise instruction following, nuanced instructions, advanced concepts, and providing speculative or incorrect answers in challenging scenarios.

Technical specifications

Context length: 256k tokens
Max output: —
Architecture: Transformer
Modalities: text->text
Parameters: ~111B
Open weights: Yes
Licence: CC-BY-NC-4.0
Regions: USA
Vendor: Cohere
Provider model key: cohere/command-a-03-2025