Command A+ 218B-A25B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
Cohere's open model for agents, multilingual and reasoning-heavy enterprise work, with image input: 218B parameters, 25B active per token (128 experts, 8 active), 128K tokens in and 64K out. Apache-2.0. Cohere publishes BF16, FP8 and a W4A4 checkpoint (132 GB) meant for two 80 GB GPUs.
Command A+ 218B-A25B needs about 133 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Includes the KV cache at 8K context, so it reads higher than the headline figure.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 73.4 GB | — | Too big | |
| Q3_K_M 3.41 bpw | 94.8 GB | — | Too big | |
| Q4_K_M 4.83 bpw | 133.6 GB | — | Too big | |
| Q5_K_M 5.67 bpw | 156.6 GB | — | Too big | |
| Q6_K 6.56 bpw | 180.9 GB | — | Too big | |
| Q8_0 8.50 bpw | 233.9 GB | — | Too big | |
| F16 16.00 bpw | 439 GB | — | Too big |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. Command A+ 218B-A25B VRAM calculator →
Get Command A+ 218B-A25B running
The cheapest catalogued GPU that runs Command A+ 218B-A25B is the Apple M3 Ultra (512 GB).
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run Command A+ 218B-A25B
Install Ollama, then run:
ollama run command-aSpecifications
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- Parameters
- 218 Billion (25B active)
- Context window
- 128K input / 64K output
- Architecture
- Mixture-of-Experts (128 experts, 8 active), 3:1 sliding-window (4,096) to full attention, SigLIP vision encoder
- Provider
- Cohere
- Licence
- Apache-2.0
- Availability
- Weights downloadable
- Practical deployment
- Workstation (large unified memory or pro GPU)
- Specified at
- Q4_K_M
- System RAM
- 256 GB
- Record updated
- 2026-11-11
Commercial use permitted. No usage restrictions beyond attribution.
Limits and caveats
- The card states 128K input and 64K output tokens; the config allows 200,000 positions.
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Command A+ 218B-A25B — frequently asked questions
How much VRAM does Command A+ 218B-A25B need?
About 133 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does Command A+ 218B-A25B run on an RTX 4090 (24 GB)?
No. Command A+ 218B-A25B needs about 133 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.
How do I run Command A+ 218B-A25B locally?
Install Ollama and run `ollama run command-a`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.