ModelBeat
Models/DeepSeek-V3.1
All models
D

DeepSeek: DeepSeek-V3.1

deepseek-v3-1

DeepSeek-V3.1 is DeepSeek’s open-weights AI model, released Aug 21, 2025. It scores 140 on the Epoch Capabilities Index.

An updated version of DeepSeek V3.

Context
—
Input
— / 1M
Output
— / 1M
Released
Aug 21, 2025
Modalities

Benchmarks

Scores on standardized evaluations. Higher is better — and rank shows where DeepSeek-V3.1 lands among all models tracked on Model Beat.

8.0
Intelligence Index
Epoch AI
8th percentile of tracked models
14.0
Coding Index
Epoch AI
14th percentile of tracked models
4.0
Agentic Index
Epoch AI
4th percentile of tracked models

Reasoning

5 evals
GPQA Diamond73.5%

Graduate-level scientific reasoning

Humanity's Last Exam6.7%

Frontier of human expert knowledge

MMLU-Pro83.3%

Broad expert knowledge

SimpleBench40.0%

Common-sense trick questions

WeirdML38.4%

Novel ML problem-solving

Coding

2 evals
LiveCodeBench57.7%

Contamination-free coding

SciCode36.7%

Scientific research coding

Math

1 evals
AIME 2024/202549.7%

Olympiad-qualifier math

Agentic & Tools

1 evals
τ²-bench34.8%

Tool-agent-user reliability

Changelog

1
  • DeepSeek-V3.1: Humanity's Last Exam improved from 6.3% to 6.7% (+6%)benchmark · Aug 6, 2026

Compare DeepSeek-V3.1 with

Frequently asked questions

What is DeepSeek-V3.1?

An updated version of DeepSeek V3.

Who created DeepSeek-V3.1?

DeepSeek-V3.1 is an AI model developed by DeepSeek, part of the DeepSeek family, released Aug 21, 2025.

How does DeepSeek-V3.1 perform on benchmarks?

On standardized evaluations tracked by Epoch AI, DeepSeek-V3.1 scores 83.3% on MMLU-Pro, 36.7% on SciCode, 57.7% on LiveCodeBench.

Is DeepSeek-V3.1 open source?

DeepSeek-V3.1 is released with open weights (Open weights (unrestricted)), so the model can be downloaded and self-hosted.

Benchmark data from Epoch AI (CC BY) and Artificial Analysis; pricing, specs & descriptions from OpenRouter. Data updated Aug 20, 2026.