"Mistral Medium" isn't a fixed product; the current release is Mistral Medium 3.5 (v26.04), a proprietary, modified-MIT-licensed model recommended by Mistral for most tasks and coding. Pick a specific named version, not the family label. Mistral doesn't publish Medium 3.5's per-token price on its main pricing page; check your account or the API pricing docs directly for the current rate.
Other good fits
Projects that specifically want a French/EU-based model provider.
+ Pros
Read pros
"Mistral Medium" names a tier, not one fixed model. The current release is Mistral Medium 3.5 (version 26.04), under a modified MIT license, and Mistral recommends it "for most tasks and coding" — its middle option between the smaller Small line and the larger Large line. It's proprietary and API-only, unlike Mistral's Apache-2.0 open-weight models.
Pick it when…
You want Mistral's own recommended general-purpose model rather than guessing between Small and Large.
Coding is part of your workload and you want a model Mistral positions specifically for that alongside general tasks.
You want a France-based, EU-headquartered model provider as an alternative to US or Chinese labs.
− Cons
Read cons
Being a proprietary, API-only model means you can't self-host Medium 3.5 or inspect its weights the way you can with Mistral's open releases; you depend on Mistral's API availability and terms. Version churn is real: Mistral has released multiple Medium versions, and a project built against "Medium" without pinning the exact version risks silently picking up different behaviour after an update.
Pick something else when…
You need to self-host or inspect the weights.Mistral's open models (Mistral Large 3, Small 4, and the Ministral line) are Apache-2.0 and can run on your own infrastructure.
You want the widest choice of models behind one key, Mistral included.OpenRouter lets you compare Mistral Medium against other providers' models.
You want a different provider's mid-tier model to compare cost and quality against.MiniMax is a lower-cost alternative worth benchmarking side by side.
Avoid if
You're building against "Mistral Medium" as a stable name; the underlying model changes with each numbered release.
Context
Every "Mistral Medium" reference should specify the exact version (currently 3.5, v26.04); Mistral has moved through numbered releases before, and a tutorial or benchmark written against an older "Medium" won't necessarily describe the current model's behaviour or price. Confirm the version before treating any claim about "Mistral Medium" as current.
Requirements
This is a hosted API model, not something you download and run yourself; access is through Mistral's API (or a cloud partner offering it), needing only an API key and network access, not local hardware.
Licence & cost
Mistral's pricing page doesn't publish Medium 3.5's exact per-token price directly; it shows Mistral Large's rate ($0.50/1M input, $1.50/1M output tokens) as a reference point, but Medium 3.5's own current rate needs checking directly in Mistral's API pricing documentation or your account console before budgeting a project around it. There's no local hardware cost since this is a hosted API; running effort is mostly monitoring token usage and re-verifying the exact model version and price periodically, since Mistral's naming and pricing structure have both changed before.
First task
Documentation-based suggestion — not run here.
Test coding and general-task quality side by side.
Get API access to Mistral Medium 3.5 specifically (confirm the exact model identifier in your request).
Write one coding task (a small, well-specified function) and one general writing task (a short summary of a document).
Run both and check the code against a test case, and the summary against the source document.
Note the exact model identifier and date you tested, so you can compare against a future version later.
Done when the code passes your test case and the summary accurately reflects the source, with both results logged against the specific model version you used.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on Mistral's pricing page and models overview documentation, checked 23 September 2026. No API key was used and no request was sent for this review. Medium 3.5's exact per-token price wasn't published on the pages checked; verify directly with Mistral before budgeting.
Mistral AI · Sep 23, 2026 — Company overview, France-based headquarters.
Mistral pricing · Sep 23, 2026 — Medium 3.5 recommended for most tasks and coding; Large pricing shown as reference; Apache-2.0 vs. commercial license distinction for open weights.
Mistral models overview · Sep 23, 2026 — Mistral Medium 3.5 (v26.04, modified MIT) listed as proprietary alongside the open Apache-2.0 model line.
Further review notes & alternatives
Alternatives and trade-offs
Mistral open models win when you need to self-host or want Apache-2.0 licensing instead of a proprietary API.
OpenRouter wins when you want to compare Mistral Medium against other providers without separate accounts.
MiniMax wins when a lower per-token price matters more than a France-based provider.
Process text and images with a large-context multimodal model.
Use it for coding and agentic application workflows.
Access hosted inference or self-host the available weights.