Cohere Command logo
Cohere Command logo

Cohere Command

Cohere

MODEL DIRECTORY

MODEL DIRECTORY

Cohere Command

Cohere Command

Command A+ is a 218B/25B-active Apache 2.0 model built for private enterprise deployment, multilingual tools, citations, and visual documents. It is extremely fast, but its broad independent reasoning score trails newer open challengers.

CURRENT MODEL SNAPSHOT

Provider: Cohere
Current anchor: Command A+
Lifecycle: Current
Weights: Open-weight under Apache 2.0
Reviewed: 21 July 2026

Abstract blue and purple gradient

Command A+ is optimized for enterprise constraints, not benchmark theater

Cohere released Command A+ on 20 May 2026. The model has 218 billion total parameters with about 25 billion active per token, a 128,000-token context window, up to 64,000 output tokens, text and image input, and support for 48 languages. It is designed for tool use, grounded answers, citations, and private deployment.

The model can run on two H100s or one B200 according to Cohere, a materially smaller serving target than many large open challengers. Apache 2.0 weights and deployment options across private infrastructure and selected clouds make Command A+ a sovereignty-oriented product.

Cohere’s main argument is operational fit: multilingual business data, retrieval, citations, document images, and tools inside controlled infrastructure. That is a different proposition from winning a general reasoning leaderboard.

Fast and enterprise-shaped, with a modest independent index

Artificial Analysis reports an Intelligence Index score of 23, around 174 output tokens per second, and about 0.42 seconds to first token for the evaluated Command A+ endpoint. Among similarly sized open reasoning models, that broad score is below the median; the latency profile is excellent.

Cohere reports 85 on Tau2 Telecom, up from 37 for the prior comparison model, and 25 on Terminal-Bench Hard, up from 3. It also reports 63 on MMMU-Pro, 75.1 on MMMU, 80.6 on MathVista, and up to 63% higher throughput with 17% lower time to first token than Command A Reasoning.

  • The provider results align with the intended telecom, tool, and document use cases.

  • The independent score argues against treating A+ as a universal frontier replacement.

  • Official documentation specifies 128K context; use that value even if an endpoint aggregator reports a different limit.

The enterprise case: sovereignty, speed, and grounded workflows

Command A+ is a strong candidate for private retrieval, multilingual customer operations, cited document analysis, and regulated environments that value deployability. It is less suitable when the workload demands the highest general reasoning performance and can tolerate a larger, more expensive model.

How we would evaluate it

Test a multilingual retrieval-and-action workflow with scanned documents, citation requirements, tool timeouts, and strict abstention. Measure accepted outcomes, citation validity, unsupported claims, language consistency, latency, and reviewer effort. Compare private deployment with the easiest hosted alternative.

Evidence used

Beam AI support status

Under evaluation. This record does not confirm a Beam integration, private-deployment topology, or regional support commitment.

Use case 1

Private multilingual RAG

Answer across controlled enterprise sources in priority languages with mandatory citations and abstention. Score source support, language consistency, unsupported claims, latency, and reviewer correction.

Use case 2

Telecom service agent

Reproduce a Tau2-style process using the actual policy, CRM, and order tools. Measure completed cases, incorrect actions, escalation quality, retries, latency, and customer-impact severity.

Use case 3

Visual document operations

Process scanned forms, charts, and document images into a typed workflow record. Insert low-quality scans and conflicting fields, then score extraction, citations, uncertainty, and review time.

Use case 4

Sovereign deployment benchmark

Compare a two-H100 or one-B200 private setup with a hosted endpoint. Include artifact provenance, throughput, security controls, observability, patching, staffing, rollback, and total cost.

Related LLMs

Curated alternatives to compare before selecting a model family.

ChatGPT

ChatGPT is OpenAI’s conversational product; enterprise automations use the underlying GPT models and APIs. This guide separates the product from the model family and explains how to evaluate the current GPT-5.6 lineup.

ChatGPT

Combine sections from a range of categories to easily assemble pages that meet the needs of your growing business.

DeepSeek

DeepSeek’s current V4 family provides Pro and Flash API models with thinking and non-thinking modes, a documented one-million-token context window, and compatible API interfaces.

DeepSeek

Combine sections from a range of categories to easily assemble pages that meet the needs of your growing business.

Perplexity

Perplexity’s Sonar API family combines language models with web search and citations. The core enterprise decision is whether provider-managed retrieval and source handling fit the workflow.

Perplexity

Combine sections from a range of categories to easily assemble pages that meet the needs of your growing business.

Gemini

Google’s Gemini family covers production and preview models for multimodal, reasoning, audio, media, and tool-oriented workloads. Model stage and endpoint stability are central to an enterprise selection.

Gemini

Combine sections from a range of categories to easily assemble pages that meet the needs of your growing business.

Claude

Anthropic’s Claude family spans frontier reasoning, coding, vision, and faster production models. The practical choice is a specific pinned model ID on the Claude API or a supported cloud platform.

Claude

Combine sections from a range of categories to easily assemble pages that meet the needs of your growing business.

Llama

Meta’s Llama 4 family provides open-weight, natively multimodal models. Scout emphasizes very long context and efficient deployment; Maverick targets stronger general multimodal capability.

Llama

Combine sections from a range of categories to easily assemble pages that meet the needs of your growing business.

Start Today

Build AI agents with the right model

See how Beam can orchestrate governed AI workflows across the model family that fits your requirements.

Start Today

Build AI agents with the right model

See how Beam can orchestrate governed AI workflows across the model family that fits your requirements.

Start Today

Build AI agents with the right model

See how Beam can orchestrate governed AI workflows across the model family that fits your requirements.

FAQs

Frequently Asked Questions

Model selection, deployment, governance, and Beam support questions answered.

What is Command A+ designed for?

It is designed for enterprise tool use, multilingual retrieval, grounded answers with citations, visual documents, and private or sovereign deployment rather than purely maximizing a general benchmark score.

How strong is Command A+ independently?

Artificial Analysis reports a score of 23, about 174 output tokens per second, and roughly 0.42 seconds to first token. That is excellent speed with a more modest broad reasoning result.

Is Command A+ open-weight?

Yes. Cohere releases the model under Apache 2.0. Verify the exact repository, model card, notices, and deployment dependencies before production use.

What is the official context window?

Cohere documents a 128K context window and up to 64K output. Use the official model specification rather than a higher value reported by an endpoint aggregator unless the provider confirms it.

Does Beam support Command A+ in production?

Beam support is currently Under evaluation. No approved private topology, integration, or Beam benchmark is attached to this record.