Mistral Devstral Medium vs Mistral Large 2

Compare Mistral Devstral Medium and Mistral Large 2, both from Mistral AI, in 8 community votes, mistral devstral medium wins 57% of head-to-head duels, tested across 18 shared challenges. Updated April 2026.

Which is better, Mistral Devstral Medium or Mistral Large 2?

Mistral Devstral Medium is the better choice overall, winning 57% of 8 blind community votes on Rival. Mistral Devstral Medium costs $0.4/M input tokens vs $8/M for Mistral Large 2. Compare their real outputs side by side below.

Key Differences Between Mistral Devstral Medium and Mistral Large 2

Mistral Devstral Medium is made by mistral while Mistral Large 2 is from mistral. On pricing, Mistral Devstral Medium costs $0.4/M input tokens vs $8/M for Mistral Large 2. In community voting, In 8 community votes, Mistral Devstral Medium wins 57% of head-to-head duels.

In 8 community votes, Mistral Devstral Medium wins 57% of head-to-head duels. Mistral Devstral Medium leads in Web Design, while Mistral Large 2 leads in Image Generation. Based on blind community voting from the Rival open dataset of 8+ human preference judgments for this pair.

Web Design: Mistral Devstral Medium wins 80% of votes
Image Generation: Mistral Large 2 wins 100% of votes
Our Verdict
Mistral Devstral Medium
Mistral Devstral MediumWinner
Mistral Large 2
Mistral Large 2Runner-up

Mistral Devstral Medium has the edge overall. In 8 blind votes, Mistral Devstral Medium wins 57% of the time.

Pick Mistral Devstral Medium for Web Design. Pick Mistral Large 2 for Image Generation. Mistral Devstral Medium is 12x cheaper per token — worth considering if cost matters.

Slight edge
Writing DNA

Style Comparison

Similarity
94%

Mistral Large 2 uses 79.6x more emoji

Mistral Devstral Medium
Mistral Large 2
50%Vocabulary43%
17wSentence Length21w
0.56Hedging0.39
4.5Bold14.4
5.3Lists7.4
0.00Emoji0.80
0.19Headings1.41
0.32Transitions0.01
Based on 21 + 10 text responses
vs

Ask them anything yourself

Mistral Devstral MediumMistral Large 2

279 AI models invented the same fake scientist.

We read every word. 250 models. 2.14 million words. This is what we found.

AI Hallucination Index 2026
Free preview13 of 58 slides
FAQ

Common questions