Mistral Medium 3.5
Open Weights
Mistral AI
Mistral's April 2026 mid-tier model: a 128B dense transformer with a 256K-token context window and strong coding performance (77.6% on SWE-bench Verified), released as open weights under a modified MIT license.
Strengths
- Strong coding: 77.6% on SWE-bench Verified
- 128B dense model with a 256K-token context window
- Open weights under a modified MIT license
- Efficient mid-tier option versus the 675B Large 3
Caveats
- Modified MIT license has some restrictions versus standard MIT
- Lower ceiling than Mistral Large 3 on the hardest tasks
- Vendor
- Mistral AI
- Context Window
- 256,000 tokens
- License
- Modified MIT
- Release Date
- 2026-04-29
- Modalities
- textimage
| Benchmark | Score |
|---|---|
| SWE-bench Verified | 77.6 |
Capabilities
Vision
Audio
Video
Tool Use
Resources
Pricing
Open weights under a modified MIT license - free to self-host; managed API/pricing via Mistral.