The Open-Source AI Stack
RSS
All models

Models · command

Command A

Open weights Cohere · 2025-03-13 · CC-BY-NC-4.0

Cohere's most-performant model at release, replacing Command R+ as the enterprise flagship. 111B parameters, 256K context, 23-language support, and 150 percent throughput vs Command R+ on two A100s or H100s. Released March 13 2025.

Cost

$2.50 / Mtok input
$10.00 / Mtok output

· as of 2026-05-21

source ↗

Speed

44.8 tok/sec output
538 ms TTFT

· as of 2026-05-21

source ↗

Architecture

tokens in Embedding vocab not disclosed × N layers Attention (not disclosed) Position encoding not disclosed context 256,000 tokens Dense MLP SwiGLU activation (standard) 111B active params Output projection tokens out
Schema-generated from data/models.yaml. Every label is auditable against the model's sources.

Specs

Architecture
dense
Total params
111B
Active params
111B
Context window
256K tokens
Attention
unknown
Position encoding
unknown
Post-training
sft, rlhf
OSI-approved
no
Data released
no
Training code
not released

Benchmarks

Each score carries the date it was published; we never infer or interpolate missing scores.

General reasoning

MMLU-Pro 71.2 as of 2026-05-21 source ↗
GPQA-Diamond 52.7 as of 2026-05-21 source ↗

Code

LiveCodeBench 28.7 as of 2026-05-21 source ↗

Math

MATH 81.9 as of 2026-05-21 source ↗
AIME 2024 9.7 as of 2026-05-21 source ↗
AIME 2025 13.0 as of 2026-05-21 source ↗

Available quantizations

GGUF llama.cpp's container; the common local format, k-quants from Q2 to Q8. runs on llama.cpp, Ollama
AWQ Activation-aware 4-bit weight quantization for GPU serving. runs on vLLM, SGLang
GPTQ Post-training 4-bit weight quantization for GPU serving. runs on vLLM, SGLang, Transformers
EXL2 ExLlamaV2's variable-bitrate format for consumer GPUs. runs on ExLlamaV2
MLX Apple MLX 4/8-bit layout for Apple silicon. runs on Apple MLX
FP8 8-bit float, frequently a native release on Hopper / Blackwell GPUs. runs on vLLM, SGLang, TensorRT-LLM
bitsandbytes On-the-fly NF4 / INT8 weight quantization inside Transformers. runs on Transformers

Verified via the Hugging Face model tree ↗. Community quantizations change over time; the families shown are those with published weights at audit time.

Notable innovations

  • · Two-GPU inference (A100 or H100) for 111B
  • · 150 percent throughput vs Command R+
  • · 23-language enterprise focus

Lineage

Replaces Command R+ as Cohere's enterprise flagship.

Derived from

Command R+ 2024-04-04

Derivatives

Sources