# Cohere vs Mistral: models, prices, licenses and deployment

> Mistral publishes its prices and open weights, while Cohere sells private deployment to enterprises and now has a Canadian government stake.

URL: https://claudegpt.si/news/cohere-vs-mistral  
Published: 2026-11-03T14:52:25.114Z  
Author: ClaudeGPT Si Newsroom  
Section: Reviews  
Topics: Cohere, Mistral AI, Language models, Enterprise AI  
Publisher: ClaudeGPT Si

## Key takeaways

- Mistral Large 4 has a 1M-token context window and costs $0.70 input and $2.10 output per million tokens, as of November 3, 2026.
- Cohere's pricing page lists no current per-token Command prices, and its Command A weights are CC-BY-NC, while Mistral's Small 4 and Ministral 3 are Apache 2.0.
- Cohere offers SaaS, hybrid cloud, Model Vault and private deployment, and Canada announced an equity investment of up to $1.6 billion in it on October 29, 2026.

Cohere vs Mistral comes down to who the customer is. Cohere sells models to enterprises and governments, with private deployment as the main offer. Mistral publishes open-weight models, with Apache 2.0 licenses on its smaller ones, and prices its API openly. This comparison is based on each company's docs, pricing pages and published results, read on November 3, 2026, and it contains no hands-on testing.

If you want to download and run a model commercially, Mistral's Apache 2.0 models are the clear fit. If you want a vendor that will install models in your own cloud or data center and sign a contract, Cohere is built for that.

## Cohere vs Mistral at a glance

| | Cohere | Mistral |
| --- | --- | --- |
| Main current models | Command A+ (May 2026), Command A, Command A Reasoning, Command A Vision, Command R7B | Mistral Large 4, Mistral Small 4, Ministral 3, Codestral |
| Context window | 128k to 256k tokens, depending on the model | 1M tokens for Mistral Large 4 |
| Listed API price | Current Command prices not on the pricing page; legacy Command R+ (08-2024) $2.50 in, $10 out per million tokens | Large 4 $0.70 in, $2.10 out; Small 4 $0.15 in, $0.60 out per million tokens |
| Open weights | Command A weights on Hugging Face under CC-BY-NC | Apache 2.0 for Small 4 and Ministral 3; Mistral Ecosystem Open Weight License for Large 4 |
| Deployment | SaaS, public or hybrid cloud, Model Vault, private (VPC or on-premises) | Hosted API, self-hosting of open weights, custom enterprise deployments |
| Focus | Enterprise and government | Open models, developers and enterprise |

Prices are in US dollars, as listed on the companies' pages on November 3, 2026, and they change.

## Models and context windows

Cohere's [models page](https://docs.cohere.com/docs/models) lists Command A+ (`command-a-plus-05-2026`) with a 128k context window and 64k maximum output. Command A (March 2025) has a 256k context and 8k output. Command A Reasoning also has 256k context and 32k output. Command A Vision and Command R7B have 128k.

Mistral's [Mistral Large 4](https://docs.mistral.ai/models/mistral-large-4) has a 1M-token context window. It is a mixture-of-experts model, which runs only part of its network for each token, with 52 billion active and 1.05 trillion total parameters. The docs list it as Public Preview as of October 6, 2026. Mistral's overview page does not list context windows for Small 4 or Ministral 3.

On context length alone, Mistral's flagship is four times larger than Cohere's biggest listed window. Neither company's docs show independent benchmark results in the pages we read, so claims about which model is "smarter" should wait for third-party leaderboards.

## API prices

Mistral's [pricing page](https://docs.mistral.ai/inference/pricing) lists per-token prices as of November 3, 2026:

- Mistral Large 4: $0.70 per million input tokens, $2.10 output, $0.07 for cached input.
- Mistral Small 4: $0.15 input, $0.60 output.
- Ministral 3 3B: $0.10 input and output.
- Codestral: $0.30 input, $0.90 output.

Cohere's [pricing page](https://cohere.com/pricing) does not list current per-token prices for Command A or Command A+. It shows legacy models for existing customers: Command R+ (08-2024) at $2.50 input and $10 output per million tokens, and Command R (03-2024) at $0.50 and $1.50. Rerank is priced per search unit (one query with up to 100 documents), with no dollar amount listed. Embed 5 and Rerank 5 appear as dedicated Model Vault instances, from $3 an hour or $2,000 a month for a small Embed instance. For current Command prices you have to contact Cohere or read the docs for your account.

So the price comparison is incomplete by Cohere's choice. What can be said is that Mistral publishes its prices and Cohere steers buyers to sales and dedicated instances.

## Licensing and open weights

Cohere's Command A model card on [Hugging Face](https://huggingface.co/CohereLabs/c4ai-command-a-03-2025) shows a 111-billion-parameter model under a CC-BY-NC license, plus Cohere Labs' acceptable use policy. NC means non-commercial, so the weights are for research and personal use. Companies that want to run it in production deploy it through Cohere.

Mistral's [models overview](https://docs.mistral.ai/getting-started/models/models_overview/) uses several licenses. Small 4 and Ministral 3 are Apache 2.0, which allows commercial use. Large 4 uses the Mistral Ecosystem Open Weight License, whose terms we did not read, so check them before building a product on it. Models Mistral labels "Premier", such as Codestral and the OCR models, have no open license listed.

## Deployment and enterprise focus

Cohere's [deployment page](https://cohere.com/deployment-options) lists four options:

- Private: on-premises or an isolated VPC, for Command, Rerank, Embed, North and Compass.
- Model Vault: a dedicated, managed inference platform.
- Public or hybrid cloud: private infrastructure combined with a public cloud.
- SaaS: managed, priced per token, which Cohere recommends for small and midsize businesses with non-sensitive data.

Command A is also listed on Oracle, with Azure and SageMaker entries in the docs, and Command A+ on Azure AI Foundry.

Mistral's [pricing page](https://mistral.ai/pricing) describes self-hosting anywhere, and custom enterprise deployments in a private cloud or on-premises. It also sells consumer and team plans: Pro at $14.99 a month and Team at $24.99 per user a month, as of November 3, 2026. Cohere's pricing page has no comparable consumer plan.

## Cohere's Canadian government stake

On October 29, 2026, the Government of Canada [announced](https://www.pm.gc.ca/en/news/news-releases/2026/10/29/canada-announces-landmark-investment-cohere-scale-secure-sovereign-ai) an equity investment of up to $1.6 billion in Cohere, made in stages. The release says Cohere committed to at least 1,000 new technical jobs in Canada by 2030, at least $1.7 billion in additional Canadian computing infrastructure by 2030, and to keeping its Canadian incorporation and headquarters. It names NVIDIA, Bell and Schwarz Group among private investors.

For buyers, the deal reinforces Cohere's pitch as a sovereign, government-friendly supplier. It does not change the models or prices above.

## Which to pick

Choose Mistral if you want open weights you can run commercially, a published price list and the largest context window. Choose Cohere if you need a vendor to deploy and support models inside your own environment. If you need current Cohere token prices, ask for a quote, because the pricing page does not give them.

## Frequently asked questions

### Is Mistral cheaper than Cohere?

On listed prices, yes in the sense that Mistral publishes them: Mistral Small 4 costs $0.15 input and $0.60 output per million tokens. Cohere's pricing page shows no current Command per-token prices, only legacy models such as Command R+ (08-2024) at $2.50 and $10.

### Are Cohere models open source?

Cohere publishes Command A weights on Hugging Face under a CC-BY-NC license, which bars commercial use. Production use goes through Cohere's API or deployment options.

### Are Mistral models open weights?

Many are. Mistral Small 4 and Ministral 3 are Apache 2.0, and Mistral Large 4 uses the Mistral Ecosystem Open Weight License. Models labeled Premier, such as Codestral, have no open license listed.

### Can I run Cohere models in my own cloud?

Yes. Cohere lists private deployment on-premises or in an isolated VPC, public or hybrid cloud, and a managed Model Vault option.

## Sources

- [Cohere models](https://docs.cohere.com/docs/models), Cohere
- [Cohere pricing](https://cohere.com/pricing), Cohere
- [Cohere deployment options](https://cohere.com/deployment-options), Cohere
- [c4ai-command-a-03-2025](https://huggingface.co/CohereLabs/c4ai-command-a-03-2025), Cohere Labs on Hugging Face
- [Mistral models overview](https://docs.mistral.ai/getting-started/models/models_overview/), Mistral AI
- [Mistral Large 4](https://docs.mistral.ai/models/mistral-large-4), Mistral AI
- [Mistral API pricing](https://docs.mistral.ai/inference/pricing), Mistral AI
- [Mistral pricing and plans](https://mistral.ai/pricing), Mistral AI
- [Canada announces landmark investment in Cohere to scale secure, sovereign AI](https://www.pm.gc.ca/en/news/news-releases/2026/10/29/canada-announces-landmark-investment-cohere-scale-secure-sovereign-ai), Prime Minister of Canada