DeepSeek V3.1

DeepSeek-V3.1 is a hybrid model that supports both thinking mode and non-thinking mode. It does better at tool calling and agent tasks, and has higher thinking efficiency than its predecessor. This is the non-thinking version. Quantized at FP8.

Native PDF input

Added Jul 26, 2025

Model weights

Pricing

Auto routing · per 1M tokens

Input: $0.20
Output: $0.70

Compare provider prices

Specifications

Context window: 128K
Max output: 65.5K
Parameters: 671B / 37B; Total / active
Avg output (7d): 324 tokens; Longer than 27% of models

Benchmarks

Artificial Analysis

LMArena

Vectara

Sourced from Artificial Analysis.

Intelligence Index

13.7

Better than 57% of models compared

Agentic work

T²-Bench Telecom (legacy)

Legacy fallback · Conversational AI agents in dual-control scenarios

34.8%

Better than 43% of models compared

Document reasoning

AA-LCR v1.1

Long context reasoning with updated grading

47.0%

Better than 42% of models compared

Reasoning

HLE

Humanity's Last Exam

6.7%

Better than 40% of models compared

IFBench

Instruction-following benchmark

37.8%

Better than 32% of models compared

Coding

Terminal-Bench Hard (legacy)

Legacy fallback · Agentic coding and terminal use

24.2%

Better than 64% of models compared

LiveCodeBench

Contamination-free coding benchmark

57.7%

Better than 64% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

49.7%

Better than 48% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

83.3%

Better than 84% of models compared

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

73.5%

Better than 54% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation