TL;DR: Mistral is France's model family, split between open weights and a paid API. The flagship is Mistral Large 3 at $2 input and $6 output per million tokens. Cheaper open models like Small, Nemo, and Ministral ship under Apache 2.0, so you can self-host them for free. There are also models for coding, vision, reasoning, and edge devices.
Mistral has the widest niche coverage of any European lab, and two ways to use it. The API on La Plateforme runs the full range, from the flagship Mistral Large 3 down to the tiny Ministral edge models. Many models are also open weight under the Apache 2.0 license, so you can download and self-host them for free. On top of the general models sit specialists: Codestral for code, Pixtral for vision, Magistral for reasoning, and more.
The numbers in this guide come from Mistral’s own model and pricing pages, plus the model cards on Hugging Face, read in July 2026. API prices are per one million tokens, input first, then output. We cover the models, the license, and how to pick. We do not cover fine-tuning recipes or the Le Chat consumer plans, which are separate topics.
Why does this matter now. Mistral is the main European option, with data residency in the EU and open weights you can run yourself. That matters for teams with strict privacy or sovereignty rules. This guide helps your team pick the right model and the right way to run it, the same way we match engineers to roles.
Key takeaways
- Two ways to run Mistral: the paid La Plateforme API, or free Apache 2.0 open weights you self-host.
- Mistral Large 3 is the flagship at $2/$6 per million tokens. Ministral 3B is cheapest at $0.04.
- Specialists cover every niche: Codestral for code, Pixtral for vision, Magistral for reasoning.
- Small, Nemo, and Ministral 8B are open weight, so you can run them with no per-token fee.
- Mistral is the main European lab, with EU data residency for teams with privacy rules.
The Mistral models at a glance
Here is the main lineup, with the model name your engineers will call. Use this table as your reference. Everything below explains the details. Prices are input then output, per one million tokens.
| Model | Context | Price in / out (per 1M) | Open? | Best for |
|---|---|---|---|---|
| Mistral Large 3 | 128K | $2 / $6 | API | Flagship, hardest tasks and agents |
| Magistral Medium | 128K | $2 / $5 | API | Step-by-step reasoning |
| Mistral Medium 3.5 | 256K | $1.50 / $7.50 | API | Frontier-class, agentic and coding |
| Mistral Small 4 | 128K | $0.15 / $0.60 | Apache 2.0 | Cheap, open, self-host |
| Codestral | 256K | $0.30 / $0.90 | API | Coding and autocomplete |
| Ministral 3B / 8B | 128K | $0.04 / $0.10 | Edge / open | On-device and cheap volume |
| Mistral Nemo | 128K | $0.02 / $0.04 | Apache 2.0 | Budget open workhorse |
Which Mistral model fits your team?
Pick your main use case to see the right model.
For high-volume or privacy-sensitive work, the open Apache 2.0 models are free to self-host, and Ministral 3B on the API starts at $0.04 per million tokens. Self-hosting needs deployment skill. Hire an AI model deployment engineer who runs open models in production.
For coding and autocomplete, Codestral is tuned for code with a 256K window at $0.30/$0.90. Pair it with engineers who ship daily. See what building this in-house costs on our developer rate card.
For reasoning and agent work, Magistral thinks step by step and Large 3 is the flagship. Strong AI and machine learning engineers get the most from them.
The model is the easy part. The hard part is the team that builds and ships with it. We match you with pre-vetted engineers in about 24 hours. Tell us what you need.
The Mistral families in plain English
Mistral splits its lineup by size and by job. The general models run from the Large flagship down to the tiny Ministral edge models. The specialists handle code, vision, and reasoning. Once you see the two axes, size and job, the long list makes sense.
Large, Medium, Small, and Ministral
Mistral Large 3 is the flagship, for the hardest reasoning and agent work, at $2 input and $6 output per million tokens. Mistral Medium 3.5 is a newer frontier-class model at $1.50/$7.50, strong on agentic and coding work. Mistral Small 4 is a cheap open model around $0.15/$0.60, and Ministral 3B and 8B are edge models small enough to run on a laptop or phone. The small and edge models are where cost-sensitive teams spend most of their traffic.
Codestral, Pixtral, and Magistral
The specialists are what make Mistral feel complete. Codestral is the coding model, tuned for autocomplete and code generation with a 256K window. Pixtral is the vision family, which reads images, charts, and documents alongside text. Magistral is the reasoning line, which thinks step by step before it answers, with a smaller open version and a stronger API version. There are also Voxtral for audio and Mistral OCR for document extraction.
Open weights and the API
Mistral is unusual in shipping both. Several models, including Mistral Small, Nemo, Ministral 8B, Mixtral, and Codestral Mamba, are open weight under the Apache 2.0 license. You can download them from Hugging Face and run them on your own hardware for free, for any use. The flagship Large, Medium, Magistral Medium, and Pixtral Large are served through the API, with some available under a research license.
The choice is the same as with other open families. Self-host the open models for full data control and zero per-token cost. Use the API for the flagship and to skip the infra. Many teams mix both, running open models for volume and calling the API for the hardest work. Running the open models well is a job for strong AI and machine learning engineers.
Pricing compared across the lineup
Mistral spans a huge price range. Output tokens drive most cost, and the range runs from $7.50 per million on Medium 3.5 and $6 on the Large flagship down to $0.04 on Ministral 3B, well over a hundred times cheaper. That spread is the point: you match the model to the job and only pay top rates when you need them. The chart shows output price across the main models.
The best way to save is to route by difficulty and self-host the open models for steady volume. Send simple work to Small or Ministral and keep only the hardest calls on Large 3 or Magistral. Batch processing runs non-urgent jobs at half price. Teams that build this routing well cut their bill by half or more. This is the kind of work that separates a strong AI engineer from an average one.
Context windows and reasoning
Most Mistral models offer a 128K token context window, roughly 96,000 words, which is plenty for chat, documents, and most coding. Codestral stretches to 256K for larger code sessions. These windows are smaller than the 1M some rivals offer, but they cover the vast majority of real workloads and keep serving cost down.
The Magistral line adds explicit reasoning. It thinks step by step before answering, which helps on hard math, code, and logic, at the cost of more output tokens. For simple tasks, a general model without reasoning is faster and cheaper. An engineer who knows when to reach for Magistral runs Mistral far cheaper at the same quality. You can read more in our breakdown of Claude versus ChatGPT for coding.
Which model should you use
Start with Mistral Small 4 as the cheap, open default for most work, or Ministral for on-device tasks. Move up to Mistral Medium 3.5 or Large 3 when the task is hard, and Magistral when it needs step-by-step reasoning. Use Codestral for code and Pixtral for vision. The table below maps common jobs to a model.
| Job | Recommended model | Why |
|---|---|---|
| Support chatbot, FAQ | Mistral Small 4 | Cheap, open, easy to self-host |
| On-device or edge | Ministral 3B or 8B | Small enough to run locally |
| Daily coding, autocomplete | Codestral | Tuned for code, 256K context |
| Harder production tasks | Mistral Medium 3.5 | Frontier-class, agentic and coding |
| Math, logic, planning | Magistral | Step-by-step reasoning |
| Images and documents | Pixtral | Native vision input |
A simple rule helps. If a cheaper or open model gives the right answer most of the time, use it and add a check. Escalate only the cases that fail. This keeps quality high and cost low. We worked with a European team that ran everything on a frontier US API. Moving routine work to a self-hosted Mistral Small model kept their data in the EU and cut their bill by more than half.
Older Mistral models
Mistral ships fast, and older versions are still around. If your code calls an older model, it will usually keep working, but the newer versions are stronger and often cheaper. The table shows the main older models and what to use instead.
| Older model | Status | Move to |
|---|---|---|
| Mistral Large 2 | Superseded | Mistral Large 3 |
| Mistral Small 3, Medium 3 | Superseded | Small 4, Medium 3.5 |
| Mixtral 8x7B, 8x22B | Legacy, open | Mistral Small 4 or Medium 3.5 |
| Mistral 7B | Legacy, open | Ministral 8B or Nemo |
A move is usually small. On the API you change the model name. For open weights you download the new checkpoint and update your serving config. The Mistral models page and the model cards on Hugging Face show every current and older version in one place, so check them before you migrate.
What this means for hiring
Knowing the lineup is step one. Getting value from it is the harder step. With Mistral the biggest levers are routing across the wide price range, self-hosting the open models, and using the specialists for code and vision. The teams that win do this well, not the ones on the biggest model. That is engineering, not just a model choice.
Engineers who can deploy and optimize open models are scarce and costly in the United States. A senior AI engineer there runs $12,000 to $18,000 per month all in. The same skill from a pre-vetted engineer in Vietnam or the Philippines often costs $3,000 to $6,000 per month. The work is the same. The bill is not. Many of our clients build their whole Mistral stack with an AI model deployment engineer from our network.
If you want to go deeper on the wider picture, our library of articles on AI tools and trends covers adoption, cost, and the open versus closed model debate. Read that next to see how fast this market is moving and why the right team matters more each quarter.
Compare other model families
Build your Mistral stack with the right team
Mistral gives you open weights, EU data residency, and a model for every niche. The model is the easy part. The team that deploys and ships with it is what creates value. We match you with pre-vetted senior engineers across Asia in about 24 hours, with no upfront cost and payroll handled.
