Skip to content
Mistral AI

Mistral Medium 3.5

Frontier

Mistral AI•Released on 2026-04-29

Mistral's unified flagship merging chat, reasoning and coding into one 128B dense model. 256K context, configurable reasoning effort (none/high), native function calling, multimodal (text+image input). Modified MIT open weights. New default in Le Chat & Vibe; replaces Medium 3.1, Magistral and Devstral 2.

84
Overall Score

Core Specs

262K
Context Window
33K
Max Output
ReasoningOpen Sourcetextimage

Scenario Scores

Coding
#27 / 5585
Writing
#21 / 5582
Translation
#17 / 5587
Data Analysis
#20 / 5584
Conversation
#26 / 5584
Image Gen
—

Pros & Cons

Sentiment0% +0% ·0% −

Pros

  • +Open-weight 128B dense model (Modified MIT)
  • +256K context window
  • +Unified chat + reasoning + coding in single weights
  • +Configurable reasoning effort per request
  • +Native function calling and JSON output
  • +Strong multilingual (24+ languages)

Cons

  • −~4x price increase vs Medium 3 ($0.4/$2 → $1.5/$7.5)
  • −Behind GPT-5.5 / Claude Opus 4.7 on top reasoning benchmarks
  • −License has commercial-revenue exception (not pure MIT)

Pricing

Input (per 1M tokens)$1.50
Output (per 1M tokens)$7.50
Free trial available
Updated on 2026-05-03

Get Started

1Visit the provider's website
2Create an account
3Start using the model

Benchmarks

mmlu89%
humanEval93%
swebench62%

Reliability

Incidents (30d)0
开源模型，可靠性取决于托管方；Mistral 官方 API 与 Le Chat / Vibe 默认提供

Compare with Others

Mistral Medium 3.5 vs Claude Opus 4.7 →Mistral Medium 3.5 vs Claude Opus 4.6 →Mistral Medium 3.5 vs Claude Opus 4.5 →