Skip to content
EN ES
Mistral AI

Mistral Medium 3.5

Released 2026-04 · hybrid · 256K tokens · 2 benchmarks

Editorial notes

Lanzado 29 abril 2026. 128B dense merged que unifica chat + reasoning + coding en un solo modelo. Reemplaza posicionalmente a Devstral 2 y Magistral Medium. Vision encoder entrenado from scratch. Self-hosting con 4 GPUs. Modified MIT. Pricing $1.50/$7.50 per MTok. SWE-bench-Verified 77.6 y AIME-2025 86.3 (avg@16, "maximum reasoning settings" — condicion distinta al canonico sin-tools) del model card oficial mistralai/Mistral-Medium-3.5-128B. Mistral NO publica GPQA-Diamond/MMLU-Pro/HLE/LiveCodeBench/SWE-bench-Pro para este modelo.

Spec sheet

Company
Mistral AI
Country
FR
Type
hybrid
Release
2026-04
Context
256K tokens
Slug
mistral-medium-3-5

Benchmarks (2)

Coding 1

Math 1

Cite this model
BibTeX · APA

BibTeX 

@misc{frontier-mistral-medium-3-5,
  title  = {Mistral Medium 3.5},
  author = {{Mistral AI}},
  year   = {2026},
  note   = {Frontier Benchmarks AI atlas. Accessed 2026-06-17},
  url    = {https://frontierbenchmarks.com/models/mistral-medium-3-5}
}

APA

Mistral AI (2026). Mistral Medium 3.5 [Large language model]. Frontier Benchmarks AI. Retrieved 2026-06-17, from https://frontierbenchmarks.com/models/mistral-medium-3-5

Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.

Battle vs another model ← All models More from Mistral AI Methodology