Back to the index
Mistral AI · Mistral
No. 21

Mistral Medium 3.5

Efficient dense workhorse with a built-in coding agent.

BalancedClosed weightsText, Image, Pdf
Field notes
Context window
262Ktokens
Max output
66Ktokens
Input price
$1.50/ 1M tokens
Output price
$7.50/ 1M tokens
Cached input
$0.15/ 1M tokens
Modalities
Text, Image, Pdf
Knowledge cutoff
—
Released
Apr 2026
Approx. speed
—
Overview

What this model is

Mistral Medium 3.5 is a proprietary dense ~128B model tuned for the price/performance sweet spot, bundling instruction-following, reasoning, and an agentic coding assistant that can open PRs. It delivers frontier-adjacent quality on many enterprise tasks at a fraction of flagship cost, but its agentic coding — while genuinely useful — is not as robust over long sessions as Claude or GPT. A strong default for cost-conscious production deployments.

Strengths

  • Excellent price/performance for enterprise use
  • Built-in agentic coding assistant
  • Strong multilingual and writing
  • 256K context with vision + PDF input

Trade-offs

  • Closed weights (API-only)
  • High output token price relative to input
  • Long-session agentic reliability below the frontier
  • Not the top pick for hardest reasoning

Best for

  • Cost-efficient production assistants
  • Multilingual enterprise workloads
  • Moderate coding automation
  • Document + vision tasks

Not ideal for

  • Fully autonomous long coding agents
  • Frontier math competitions
Capabilities

Capability profile

Normalized 0–100 scores, comparable across the whole catalog.

Reasoning73Coding71Math71Writing77Knowledge78Speed74Agentic68Vision68Multilingual84Long Context76
Reasoning
73
Coding
71
Math
71
Writing
77
Knowledge
78
Speed
74
Agentic
68
Vision
68
Multilingual
84
Long Context
76
Benchmarks

How it scores

Public benchmark results, with independent third-party results where available. Bars normalize percentages to 100 and Elo ratings to a 1500 ceiling.

Independent index
Artificial Analysis Intelligence Index
Composite of ~9–10 independent evals · Artificial Analysis
30/ 100
BenchmarkResult
Artificial Analysis Intelligence Indexindependent
GeneralArtificial Analysis
30%
SWE-bench Verifiedindependent
CodingSWE-bench
77.6%
Alternatives

Comparable models

Models in a roughly similar class — worth weighing against this one.

Mistral Large 3Mistral AI

Europe's flagship open-weight MoE — a strong, cheap, multilingual generalist.

Qwen3.7-MaxAlibaba Qwen

Alibaba's proprietary agentic reasoning flagship with a 1M context.

GPT-5.4 miniOpenAI

Affordable OpenAI workhorse for standard production routing.