Back to models
NewChat Models

Cohere

Command A+

Best for Research

Multimodal Cohere MoE with hybrid reasoning, 48-language support and a 128K context window.

Vision48 languages
Open AppBrowse All Models

At a glance

Know the model before you prompt.

Cohere API specifications
Context window
128,000 tokens
Maximum output
64,000 tokens
Inputs → output
Text + Images → Text
Knowledge cutoff
April 1, 2025

API model ID: command-a-plus-05-2026

Capabilities & boundaries

What it supports. Where the limits are.

Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.

Supported API features and tools

  • A mixture-of-experts design with 25B active and 218B total parameters
  • Text and image input with text output across 48 supported languages
  • Hybrid thinking for reasoning-intensive tasks, enabled by default
  • Tool use and structured outputs listed on the A+ card for configured integrations

Before you choose

  • Image understanding does not imply native image, audio or video generation. Check visual claims against the original.
  • The model card's 128,000 context and 64,000 output limits are distinct. Thinking and final-answer budgets must be planned together.
  • Free API use is bounded by rate limits; production deployment is directed to Model Vault, not an unlimited zero-cost endpoint.
  • Context capacity is not a guarantee of complete recall. Keep source IDs, evaluate omissions and verify citations against the underlying documents.
  • A generated plan or tool call does not authorize or execute an external action. Use a configured integration, validate results and keep consequential changes behind human approval.

Extended thinking

enabled · defaultdisabled

Cohere's hybrid reasoning uses the thinking object and is enabled by default. Set type disabled to turn it off. An optional token_budget limits thinking; leave room for the final response, with the provider recommending at least 1K answer tokens. Do not mistake the guide's 31K example for a universal limit across all reasoning models.

  • Use command-a-plus-05-2026 for the model covered here. The website slug is a navigation label, not necessarily the API model ID.
  • Use the model's thinking schema, not another provider's reasoning_effort enum. Parse thinking blocks separately from the final text.
  • The model card lists text/image inputs, tool use and structured outputs. Those capabilities still require integration configuration and validation; they do not grant automatic data access.
  • Evaluate language-specific quality using native reviewers and domain terminology rather than treating the number of supported languages as equal accuracy in every language.

Put it to work

Start with a more useful prompt.

Original examples from EZ Ai Assist. Adapt these to your task and the features available in your workspace.

Workflow 01

Compare multilingual visual evidence

Keep observations traceable across languages.

Compare these supplied product screenshots and accompanying text in their original languages. Build a table of visible facts, translated meaning, source image and uncertainty. Preserve names and numbers exactly. Flag ambiguous wording for a native speaker rather than guessing. End with the discrepancies we should verify before publishing a unified product description.

Workflow 02

Plan a controlled document workflow

Separate useful automation from actions needing approval.

Design a workflow for the supplied document-review task and available tools. Identify what evidence each step needs, how tool results should be validated, and where a person must approve changes. Include a recovery path for missing or conflicting information. Do not execute actions or claim access to systems that are only mentioned in the description.

Workflow 03

Evaluate structured extraction quality

Test correctness as well as valid formatting.

Use this field specification and sample documents to propose an extraction evaluation. Include valid examples, unreadable values, conflicting facts and missing required fields. Define checks for both schema compliance and source accuracy. Return a test matrix with expected outcomes; do not invent values for documents that do not state them.

Developer reference

Cohere API pricing

These are Cohere API reference terms, not EZ Ai Assist subscription prices.

View EZ Ai Assist plans

The current model card offers trial and production keys free usage until rate limits are reached. This is not unlimited free production. Cohere directs production deployments to Model Vault. No public per-token production rate is established here.

  • Confirm limits for your exact model and key type; a production key alone is not evidence of unrestricted capacity.
  • Obtain deployment-specific commercial terms before budgeting. Do not substitute another Command model's token rates or show unverified prices as $0.

Common questions

A few things worth knowing.

What distinguishes A+ from Command A?

A+ adds a mixture-of-experts architecture, vision and hybrid reasoning, with 48-language support. Its published context is 128,000, not Command A's 256,000; newer does not mean every limit is larger.

Can thinking be disabled?

Yes. Cohere's reasoning models default to enabled thinking and support type disabled. An optional token budget controls thinking; reserve space for the final answer and test the tradeoff on real tasks.

Is A+ free for unrestricted production?

No. The card's free usage is subject to rate limits for both key types. It directs production use to Model Vault; obtain terms for your deployment rather than treating missing public rates as zero.

Does vision mean it creates images?

No. A+ accepts images alongside text and produces text. Use it for interpretation and extraction, with a visual check of important details, not as a native image-generation model.

Can it see current information automatically?

No. Model knowledge and a long context are not a live search service. Provide current evidence or connect an authorized retrieval workflow where supported, and verify the sources before acting on an answer.

How should I test it on my own work?

Use representative examples, explicit pass criteria and difficult counterexamples. Score factual support, omitted requirements and invalid outputs as well as latency and cost. Keep a human escalation path for uncertain or consequential results.

Are these API capabilities included in my EZ Ai Assist plan?

This guide describes the provider API, not subscription entitlements or a promise that every setting is exposed in the app. Check your workspace for model access and supported inputs, and use the EZ Ai Assist pricing page for plan details.

Check the source

Official documentation

Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.

Same provider

More from Cohere

Cohere

Command A

Text-focused Cohere model for grounded answers and tool workflows, with 256K context and 8K output.

Cohere

Command A Reasoning

Hybrid text reasoning with optional thinking budgets, 256K context and 32K output for evidence-heavy workflows.

Cohere

Command A Vision

Text and image understanding for documents, tables and charts, with up to 20 images per request and no tool use.

Cohere

Command A Translate

Specialized text translation across 23 languages, with separate 8K input and 8K output limits.