Command A — local AI model by Cohere
Written by Jakub Rusinowski · Last updated
Cohere's Command A line for agents, retrieval with citations and multilingual work. Command A+ (May 2026) is an Apache-2.0 mixture-of-experts with image input, unlike the non-commercial Command R models.
Variants
The smallest Command A variant needs about 133 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| Command A+ 218B-A25B → 218B (25B active) | ~132.9 GB | 128K input / 64K output | ollama run command-a |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run Command A locally
Install Ollama, then pull the tag.
ollama run command-aPick a size above for its own VRAM figure, speed estimate and install command.
Licence
Commercial use permitted. No usage restrictions beyond attribution.
Applies to: Command A+ 218B-A25BRecommended GPU
The cheapest catalogued GPU that runs Command A locally (min 133 GB VRAM) is the Apple M3 Ultra (512 GB).
Command A — frequently asked questions
How much VRAM does Command A need?
Command A needs about 133 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Command A+ 218B-A25B (133 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Command A on an RTX 4090 (24 GB)?
Command A's smallest variant needs about 133 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
What quantization should I use for Command A?
Q4_K_M is the best balance of quality and VRAM for Command A in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Command A with Ollama?
Command A has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.