Mistral Medium 3.5
A dense 128 billion parameter model designed to run on as few as four GPUs, which makes self hosting realistic rather than theoretical.
The facts at a glance
Pricing last checked August 7, 2026. Prices change and vary by region. Check the official pricing page ↗
Assessment
What it does well
- Mistral reports 77.6 percent on SWE-Bench Verified and 91.4 on tau3-Telecom
- Targets self hosting on as few as four GPUs, which is achievable for a small team
- It replaced Devstral 2 as the default in Mistral's own coding agents, which is a meaningful internal endorsement
Where it falls short
- A modified MIT licence is not plain MIT and needs reading
- 256,000 token context is a quarter of what the current flagships offer
- Mistral's own pages give conflicting release dates, so the exact date is unverified
Privacy and data control
Self hosting keeps data on your own infrastructure. Le Chat and Mistral Vibe are governed by Mistral's own terms, which we have not verified for this profile.
The longer read
Dense models still have a place
Most frontier releases now use mixture of experts designs with enormous total parameter counts. Medium 3.5 is dense and comparatively small, which makes it predictable to serve and realistic to run on hardware a startup can actually buy. That trade, capability ceiling for operational simplicity, is the right one more often than the benchmark tables suggest.
How it performs by outcome
We assess products against a job, not a leaderboard. These are the outcomes this entry is compared under.
Worth comparing against
How this profile was produced
Source review. Every fact here is drawn from the provider documentation linked below and checked on the date shown. We have not yet run this entry through a controlled test, and no scoring is implied.
Read the full review methodology →