Devstral 2 2512

Release Date: Dec 9, 2025     Creator: Mistral AI

Devstral 2512 is a powerful code agent model built by Mistral AI for complex software engineering tasks.

Model Specifications
Knowledge Cutoff
early 2025
Context (VividLLM)
32,000 Tokens
With a context window of 32,000 tokens, this model can 'remember' and analyze up to 48 pages of data at once. This makes it a top choice for reviewing legal contracts and dense academic chapters.
Context (Native)
262,144 Tokens
Input ModalitiesTextImagePDF
Output TypeText
Max Output8192
Input Weight
0.50 x (Casual)
Output Weight
0.50 x (Casual)
Industry Benchmarks
Intelligence Index (Artificial Analysis) N/A
Coding Index (Artificial Analysis)N/A
Math Index (Artificial Analysis)N/A
Response Speed
Output Tokens per second N/A
Median Time to First Token SecondsN/A
Launch VividLLM Now => :

Benchmarks accessed via : Artificial Analysis

Disclaimer: We are providing Image Upload options to only limit models as of now, so some models which have image upload options might now have that feature on our site.