Command A+

Overview

Command A+ is Cohere’s flagship model for enterprise agentic workflows. It accepts text and image inputs and has a 192,000-token context window, giving applications room to provide substantial conversation history or source material in a single request. The model produces text output, supports reasoning, and offers native tool calling with strict tool schemas.

For teams evaluating it as part of an existing stack, the central fit question is whether those capabilities and its token pricing suit the intended workflow. The listed model ID is cohere/command-a-plus. It was released on September 22, 2026, and is part of Cohere’s model family. Explore Cohere models, newest models, or models with tool calling.

Key features

Tool use and reasoning

Command A+ supports native tool calling and strict tool schemas, making tool use a core part of its stated agentic-workflow positioning. Reasoning is also listed as supported. These details may matter when building applications that need a model to work with defined tools and produce text responses, but the available facts do not specify particular tools, reasoning controls, or task-level performance.

Long context and image input

The 192,000-token context window is paired with both text and image input. That gives builders a large context allowance and multimodal input options; however, no image-processing details or supported image formats are specified. Outputs are text, and the model’s listed modality is text.

Output limit

Maximum output is 64,000 tokens. This is the listed ceiling, not a statement about typical response length or speed.

Pricing

Input costs USD 0.3 per 1M tokens, and output costs USD 1.5 per 1M tokens. The separate rates make output tokens five times as expensive as input tokens, so teams estimating usage should account for both sides of their workload. No other price terms are provided. For additional cost context, see cheapest language models.

Platforms

The available facts identify Command A+ as a Cohere model and give its model ID as cohere/command-a-plus. They do not specify SDKs, deployment options, API compatibility, or supported hosting platforms, so those integration details cannot be assessed here.

Who it's for

Command A+ is positioned for enterprise agentic workflows. Its native tool calling, strict tool schemas, reasoning support, large context window, and text-and-image inputs are relevant considerations for teams selecting a model for workflows that combine supplied context with tool use. It may also merit evaluation by builders needing image input alongside text output. The stated facts do not establish performance, reliability, or suitability for any particular industry or task.

Teams whose requirements center on multimodal inputs can browse vision language models, while those comparing reasoning capabilities can explore reasoning models.

Pros and cons

Pros

• 192,000-token context window.
• Text and image inputs, with text output.
• Native tool calling with strict tool schemas.
• Reasoning support and a maximum output of 64,000 tokens.

Cons

• Output tokens cost USD 1.5 per 1M tokens, compared with USD 0.3 per 1M input tokens.
• Available details do not specify deployment options, SDK support, or image-handling specifics.
• No performance or task-level results are provided for assessing how well it performs in a given workflow.

Alternatives

Within Cohere’s lineup, Command A, Command R7B (12-2024), Command R (08-2024), and Command R+ (08-2024) are listed alternatives. North Mini Code (free) is another option in the provided list. No specifications or prices for these alternatives are given here, so a direct capability or cost comparison is not possible.

Verdict

Command A+ presents a defined set of capabilities for enterprise agentic workflows: native tool calling with strict schemas, reasoning support, text and image inputs, and a 192,000-token context window. Its 64,000-token maximum output provides a further sizing parameter, while pricing distinguishes between USD 0.3 per 1M input tokens and USD 1.5 per 1M output tokens. The main gaps for a stack decision are practical integration and performance details, which are not specified. Teams can judge its fit against their own requirements using the published model ID, input and output rates, and listed capability limits.