Cohere's current Command model, Command A+, is live on the Cohere API but has no public per-token price. This page explains what Cohere does publish, the documented ways to run Command A+, and how to estimate its cost when the hosted price has to be confirmed with Cohere.
Cohere doesn't publish a metered per-token price for Command A+, even though its model docs list it as live on the Cohere API Chat endpoint (command-a-plus-05-2026) and on Azure AI Foundry. Free trial keys let you test it; for production pricing, ask Cohere. The weights are free under Apache 2.0, so the other way to price it is your own GPU bill — Cohere says the W4A4-quantised release fits on one B200 or two H100 GPUs.
| Model | Tier | Input / 1M | Output / 1M | Cache hit / 1M | Context | Notes |
|---|---|---|---|---|---|---|
| Command A+ | Specialist | — | — | — | 128K | Succeeds Command A; open-weight (Apache 2.0); live on Cohere's standard API (command-a-plus-05-2026) and Azure AI Foundry, but Cohere publishes no per-token rate; 64K max output |
“—” means Cohere doesn't publish a metered per-token price for that model.
Cohere's public pricing covers free trial keys, pay-as-you-go production keys, per-hour or per-month Model Vault instances for its Embed, Rerank and Parse models, and legacy Command and Command R models priced for existing customers. Command A+ isn't on the list. Trial keys are free but rate limited and can't be used for production or commercial work; a production key needs account Owner privileges or a Go to Production application. Model customisation and private deployments are on custom pricing.
Cohere documents three routes. On Cohere's own platform, the model is live on the Chat endpoint as command-a-plus-05-2026, and private deployments on-prem or in isolated VPCs are priced per customer. On Azure AI Foundry the model is listed as coherelabs-command-a-plus-05-2026-w4a4; Cohere's model table shows it as not available on Amazon Bedrock, SageMaker or Oracle OCI. Or you download the weights from Hugging Face under the Apache 2.0 licence, which allows commercial use, and serve them on your own hardware.
Command A+ is a sparse mixture-of-experts model with 218B total and 25B active parameters, accepts text and image input, and returns up to 64K output tokens. Cohere quotes 1× B200 or 2× H100 at W4A4 quantisation. Your effective price per token is your GPU-hour cost divided by sustained throughput, so run real traffic through a test deployment and measure tokens per hour before comparing it with metered APIs such as Mistral Large 3 ($0.50 / $1.50 per 1M) or DeepSeek V4-Pro ($0.66 / $1.98).
With no published per-token price, Command A+ fits teams that already need a private deployment inside their own network, a Cohere enterprise contract, or Azure-based procurement. You can try it free on a rate-limited trial key, but you can't forecast a production bill until Cohere quotes one. If you need a known price from day one, a metered API from the comparison table on the API hub is the faster start; you can move to self-hosted Command A+ later without licence fees.
Cohere doesn't list a per-token price on its pricing page, although Command A+ is live on the Cohere API and on Azure AI Foundry. Trial-key calls are free but rate limited; for production rates, contact Cohere or check Azure's listing. Self-hosting the Apache 2.0 weights costs only your infrastructure.
The weights are free under Apache 2.0 on Hugging Face, including for commercial use. Running them still needs GPUs — Cohere cites one B200 or two H100s for the W4A4 build.
No. Cohere's model table lists Command A+ as not available on Amazon Bedrock, SageMaker or Oracle OCI; it is available on Azure AI Foundry.
No. Trial API calls are free but rate limited and not permitted for production or commercial use; you need a pay-as-you-go production key.
Cohere states 1× B200 or 2× H100 GPUs at W4A4 quantisation for the 218B-parameter (25B active) model.
How we choose: prices come from each vendor's official pricing page and are re-checked every week. We take no affiliate fees, so rankings aren't for sale.