Model

Command A+

OPEN

Cohere · Command A family · released May 20, 2026

Successor to the CC-BY-NC-licensed Command A (Mar 2025), now released fully open-weight under Apache 2.0. First Command model with native image input, expanded from 23 to 48 languages, and large agentic/enterprise gains (tau2-bench-telecom 37%->85%, Terminal-Bench-Hard agentic coding 3%->25%).

ReasoningCodingVisionFunction callingTool useAgentic

1749.0

Elo · rank #84

Parameters

218B

Active params

25B (MoE)

Context

128K tokens

Architecture

Sparse Mixture-of-Experts (218B total / 25B active), hybrid attention, native citation grounding, runs on 1x B200 or 2x H100 (W4A4)

License

Apache 2.0

Languages

48+

API price (in/out)

No hosted API

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath90.0%#46

best: GPT-5.2 · 100.0%

CharXivVision52.7%#29

best: Muse Spark 1.1 (xhigh) · 88.4%

GPQA DiamondReasoning76.1%#97

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning11.4%#82

best: Claude Sonnet 5 · 57.4%

IFBenchReasoning73.9%#19

best: MiniMax M3 · 83.0%

LiveCodeBenchCoding79.4%#51

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MathVistaVision80.6%#13

best: Seed 2.1 Pro · 90.7%

MMLU-ProKnowledge78.2%#63

best: Claude Fable 5 · 91.5%

MMMU-ProVision63.0%#37

best: Claude Opus 4.7 · 85.5%

MMMUVision75.1%#35

best: Claude Fable 5 · 89.3%

τ²-Bench TelecomAgents85.0%#23

best: Claude Opus 4.6 · 99.3%

Terminal-Bench 2.0Coding17.6%#57

best: GPT-5.6 · 88.8%

Run it locally

VRAM @ Q4

—

VRAM @ FP16

—

Fits on (Q4)

Multi-node cluster required

Throughput data unavailable.

Quantizations

—

Command A family

Elo progression across releases

Command AMar 20251177.5 Command A+May 20261749.0

Resources

HFCohereLabs/command-a-plus-05-2026-bf16 DOCShttps://cohere.com/blog/command-a-plus APIAvailable via Cohere API

API price weights · each benchmark row carries its own source badge (see methodology)