0

DeepSeek V3.1

Open weights

DeepSeek's August 2025 mid-generation update - a single hybrid model that switches between thinking and non-thinking modes.

Publisher
DeepSeek
Params
671B total / 37B active
Context
128K
Input
Output
License
MIT
Released
21 Aug 2025

Cite

Notes

Only stored in your browser.

Attribution

Benchmark scores
AAprime-hubVals AI
Attribution policy →
Context
128K
Input
$0.59
/1M tok
Output
$1.69
/1M tok
Blended
$0.86
#301/ 414/1M · 3:1
Speed
0
#221/ 414tok/s

Price & speed via Artificial Analysis · ranked across tracked models

Reported on 12 evals across 8 domains - best 89.7% on AIME 2025: Problems from the American Invitational Mathematics Examination · #12 of 207

Reported eval scores

12