DeepSeek V3.1 Thinking

Thinking enabled version of 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. Quantized at FP8.

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): 1.2K tokens; Longer than 76% of models

Benchmarks

Artificial Analysis

LMArena

Vectara

Sourced from Artificial Analysis.

Intelligence Index

13.5

Better than 52% of models compared

Agentic work

T²-Bench Telecom (legacy)

Legacy fallback · Conversational AI agents in dual-control scenarios

37.4%

Better than 45% of models compared

Document reasoning

AA-LCR v1.1

Long context reasoning with updated grading

56.7%

Better than 49% of models compared

Reasoning

HLE

Humanity's Last Exam

14.3%

Better than 61% of models compared

IFBench

Instruction-following benchmark

41.5%

Better than 42% of models compared

Coding

Terminal-Bench Hard (legacy)

Legacy fallback · Agentic coding and terminal use

25.0%

Better than 66% of models compared

LiveCodeBench

Contamination-free coding benchmark

78.4%

Better than 90% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

89.7%

Better than 90% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

85.1%

Better than 91% of models compared

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

77.9%

Better than 63% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation