Command A Plus 05 2026 BF16 — Hardware Requirements & GPU Compatibility
VisionChatFunctions
Command A+ (command-a-plus-05-2026, BF16 checkpoint) is Cohere and Cohere Labs' large mixture-of-experts model built for agentic, multilingual, and reasoning-heavy enterprise work, with vision input support alongside text. It has 218 billion total parameters and roughly 25.5 billion active per token, routing across 128 experts (8 active per token, plus one shared expert applied to every token) and interleaving three sliding-window attention layers with one global-attention layer, a pattern first used in the original Command A. It exposes explicit step-by-step thinking before its answer. This BF16 release needs a multi-GPU server; Cohere also ships FP8 and W4A4 quantizations that shrink the hardware footprint considerably. Context length is 131,072 tokens. It is released under the Apache 2.0 license, permitting unrestricted commercial and research use, and was published in May 2026.
Command A Plus 05 2026 BF16 runs on a 192 GB+ Mac or a datacenter GPU
Best local pick → Mac Pro M2 Ultra (192 GB) · Q4_K_M · 132.1 GB · ~30 tok/s
Macs from 96 GB (IQ3_XS · ~40 tok/s)
Apache 2.0
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Command A Plus 05 2026 BF16
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
ollama run hf.co/bartowski/command-a-plus-05-2026-GGUF:Q4_K_MQ4_K_M runs on one H200 NVL 141 GB · 16K context
Download: command-a-plus-05-2026-Q4_K_M/command-a-plus-05-2026-Q4_K_M-00001-of-00004.gguf135.09 GB + 3 partsbartowski
Other ways
each opens step-by-stepRun in cloud
Fits on B200 (192 GB) (59 GB headroom) · Q4_K_M
- Generation speed
- ~212 tok/s
- generation speed
- Cost per 1M output tokens
- $7.82
- per 1M output tokens
How Much VRAM Does Command A Plus 05 2026 BF16 Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 93.3 GB | 94.4 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB | |
| 96.1 GB | 97.1 GB | 3-bit small — 3-bit small quantization | Runs on Instinct MI250X 128 GB | |
| 107.0 GB | 108 GB | 3-bit medium — 3-bit medium quantization | Runs on Instinct MI250X 128 GB | |
| 109.7 GB | 110.8 GB | 4-bit · legacy — 4-bit legacy quantization | Runs on Instinct MI250X 128 GB | |
| 131.6 GB | 132.6 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Runs on H200 NVL 141 GB | |
| 156.2 GB | 157.2 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Runs on B200 192 GB | |
| 180.8 GB | 181.9 GB | 6-bit, very good — 6-bit quantization, very good quality | Runs on B200 192 GB | |
| 219.1 GB | 220.1 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Runs on Instinct MI325X 256 GB |
Selected Q4_K_M
4-bit medium quantization132.1GBat 16K context
- Weights
- 131.3 GB
- KV cache
- 0.5 GB
- Overhead
- 0.3 GB
Server-class: ~132.1 GB at 16K context · one H200 NVL 141 GB
Which Hardware Can Run Command A Plus 05 2026 BF16?
Q4_K_M · 131.6 GBCommand A Plus 05 2026 BF16 (Q4_K_M) requires 131.6 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 172+ GB is recommended. Using the full 200K context window can add up to 6.5 GB, bringing total usage to 138.1 GB. No consumer GPU has enough memory. 4 devices with unified memory can run Command A Plus 05 2026 BF16, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (192 GB).
128–144 GB
Q4_K_M · best fit Q4_1 (123.9 GB)
4 GPUs · 2 Mac classes
GH200 Grace Hopper Superchip~159Simulate output on GH200 Grace Hopper SuperchipH200 NVL~157Simulate output on H200 NVLH200 SXM~157Simulate output on H200 SXMM2 Ultra*128 GB~31Simulate output on Mac Pro M2 Ultra (128 GB)M5 Max128 GB~25Simulate output on Mac Studio M5 Max (128 GB)192+ GB
fits · 59.9 GB free · up to 195K ctx
6 GPUs · 1 Mac class · 2 device families
B200~212Simulate output on B200B300~212Simulate output on B300Instinct MI350X~104Simulate output on Instinct MI350XM3 Ultra*256–512 GB~30Simulate output on Mac Studio (M3 Ultra, 256GB)8x H100 SXM5~379Simulate output on NVIDIA DGX H1008x A100 SXM4~312Simulate output on NVIDIA DGX A100 640GB
By VRAM tier · typical GPUs and Macs at Q4_K_M
Consumer GPUs (up to 32 GB) and Macs up to 64 GB: too small for this model (needs ≥ 61 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
How good is Command A Plus 05 2026 BF16?
llmrun Score blends public benchmark boards into one number; 100 = the average open model.
Not enough data yet for an llmrun Score.
A Score needs results on enough independent boards. How the Composite Score is built
Benchmarks
Here are all 2 boards that have results for this model, sorted by their weight in the llmrun Score; badges show who ran the test.
Similar models
4| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| MiniMax M2.7MiniMax | 107.4 | 139.6 GB | —needs > 24 GB | same size |
| Qwen3 VL 235B A22B InstructAlibaba | — | 143.3 GB | —needs > 24 GB | same size |
| DeepSeek V2.5 1210DeepSeek | — | 142.9 GB | —needs > 24 GB | same size |
| Laguna M.1poolside | — | 138.1 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- Cohere
- Family
- Command
- Parameters
- 218.8B
- Active parameters
- 25.5B per token
- Architecture
- Cohere2VisionForConditionalGeneration
- Context Length
- 200,000 tokens
- Vocabulary Size
- 262,144
- Release Date
- 2026-05-11
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Command A Plus 05 2026 BF16 need?
Command A Plus 05 2026 BF16 requires 131.6 GB of VRAM at Q4_K_M, or 437.9 GB at BF16. Full 200K context adds up to 6.5 GB (138.1 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 218.8B × 4.8 bits ÷ 8 = 131.3 GB
KV Cache + Overhead ≈ 0.3 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 6.8 GB (at full 200K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Command A Plus 05 2026 BF16?
No — Command A Plus 05 2026 BF16 requires at least 60.5 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
What's the best quantization for Command A Plus 05 2026 BF16?
For Command A Plus 05 2026 BF16, Q4_K_M (131.6 GB) offers the best balance of quality and VRAM usage. Q4_K_L (134.3 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 60.5 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Command A Plus 05 2026 BF16 on a Mac?
Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run Command A Plus 05 2026 BF16, the smallest at IQ3_XS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 131.6 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (192 GB).
Can I run Command A Plus 05 2026 BF16 locally?
Yes — Command A Plus 05 2026 BF16 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Command A Plus 05 2026 BF16?
At Q4_K_M, Command A Plus 05 2026 BF16 can reach ~104 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 15.3 GB active ÷ (8000 × 0.65) = 2.95 ms, plus 32 layers × 0.055 ms = 1.76 ms, so 1000 ÷ 4.71 ms = ~212 tok/s
Estimated speed at Q4_K_M (131.6 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Command A Plus 05 2026 BF16?
At Q4_K_M, the download is about 131.25 GB. The full-precision BF16 version is 437.50 GB. The smallest option (IQ2_XXS) is 60.16 GB.
Which GPUs can run Command A Plus 05 2026 BF16?
No single consumer GPU has enough VRAM to run Command A Plus 05 2026 BF16 at Q4_K_M (131.6 GB). Multi-GPU or professional hardware is required.
Which devices can run Command A Plus 05 2026 BF16?
4 devices with unified memory can run Command A Plus 05 2026 BF16 at Q4_K_M (131.6 GB), including Mac Pro M2 Ultra (192 GB), Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M2 Ultra (192 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.