Command A model review
cohere/command-aAn open-weights text model for agent-style, multilingual, and coding tasks.
256K tokens of context — about 192,000 words, on a scale that ends at 2M.
What it can do
- Open weights
- Takes text
- Gives text
Released 13 Mar 2025 · checked 3 Oct 2026.
About it
Command A is Cohere’s open-weights language model with 111 billion parameters. It reads up to 256,000 tokens at once and returns text, with an 8,192-token output limit. The model is aimed at agent-style work, multilingual tasks, and coding. Input costs $2.5 per million tokens, while output costs $10 per million tokens.
Review
In a line
Command A is Cohere’s open-weights model with 111 billion parameters. It takes text in and produces text out.
What it's good at
The model is intended for agent-style tasks, multilingual work, and coding. Its 111 billion parameters and large context window support work with substantial text inputs.
Command A is text-only. It does not take images as input or produce images.
Price in plain terms
Input costs $2.5 per million tokens, and output costs $10 per million tokens. Output is therefore priced higher than input.
Reading a page of about 500 words costs about 0.16 cents. A chat turn with roughly 1,000 input tokens and 500 output tokens costs about 0.75 cents. 1,000 such turns cost about $7.50.
Context and limits
Command A can read up to 256,000 tokens in one request. Tokens are small pieces of text, so this allows a large amount of material to be supplied at once.
A single answer can contain up to 8,192 tokens. The model accepts text and returns text, with no image modality listed.
Ways to use it
Command A is available as an open-weights model. Its model ID is cohere/command-a, and its Hugging Face ID is CohereForAI/c4ai-command-a-03-2025.
Verdict
Choose Command A if you need a text model for coding, multilingual requests, or agent-style workflows and want open weights. The 256,000-token context is useful when a task depends on a lot of source material.
Look elsewhere if you need image input or output, or if an 8,192-token response limit does not fit your work. Its output rate is also much higher than its input rate, so long answers can raise costs faster than long prompts.
Verdict
Command A fits teams that want open weights, broad text handling, and support for multilingual, coding, or agent-style tasks. Its 256,000-token context can hold extensive material, while the 8,192-token output cap sets a clear ceiling for each answer. Text-only operation rules out image-based work. Costs are modest for short exchanges, but generated text is priced at $10 per million tokens, four times the input rate. Consider it when those capabilities match your workload and you can manage output spending.
- Open-weights model from Cohere
- 256,000-token context window
- Designed for multilingual and coding tasks
- Input costs $2.5 per million tokens
- Text-only input and output
- Maximum answer length is 8,192 tokens
- Output costs $10 per million tokens
- Pick it ifCoding and multilingual text tasks; Agent-style workflows using large source documents; Teams seeking an open-weights model
- Skip it ifYou need image understanding or image generation; You regularly need answers longer than 8,192 tokens; Your workload produces lots of output tokens
Pricing
Input is $2.5 per million tokens and output is $10 per million tokens. A typical 1,000-token-in, 500-token-out chat turn costs about 0.75 cents; 1,000 such turns cost about $7.50.
Questions
What is Command A?
Command A is Cohere’s open-weights text model with 111 billion parameters.
How much text can Command A read?
Its context window is 256,000 tokens, meaning that is the maximum amount of text it can handle in one request.
How long can one answer be?
The maximum output is 8,192 tokens.
What does Command A cost?
Input costs $2.5 per million tokens. Output costs $10 per million tokens.
Does Command A work with images?
No. The listed input and output modality is text.
What is Command A suited to?
It is focused on agent-style, multilingual, and coding use cases.
Where can I identify the model?
Its model ID is cohere/command-a, and its Hugging Face ID is CohereForAI/c4ai-command-a-03-2025.