DeepSeek V3/Deepseek Chat

DeepSeek original V3 model, trained on nearly 15 trillion tokens, matches leading closed-source models at a far lower price. Quantized at FP8.

Native PDF input
Tool Calling
Structured Output

Added Feb 27, 2025

Model weights

Pricing

Auto routing · per 1M tokens

Input: $0.10
Output: $0.42

Compare provider prices

Specifications

Context window: 128K
Max output: 8.2K
Parameters: 671B / 37B; Total / active
Avg output (7d): 128 tokens; Longer than 9% of models

Benchmarks

Artificial Analysis

Vectara

Design Arena

Sourced from Artificial Analysis.

Intelligence Index

16.0

Better than 58% of models compared

Coding Index

21.2

Better than 25% of models compared

Agentic Index

0.8

Better than 7% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

0.3%

Better than 5% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

66 Elo

Better than 8% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

73 Elo

Better than 11% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

1.6%

Better than 12% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

45.7%

Better than 41% of models compared

Reasoning

HLE

Humanity's Last Exam

11.2%

Better than 54% of models compared

IFBench

Instruction-following benchmark

49.0%

Better than 58% of models compared

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

0.0%

Better than 16% of models compared

SciCode

Python programming for scientific computing

39.0%

Better than 23% of models compared

LiveCodeBench

Contamination-free coding benchmark

59.3%

Better than 66% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

59.0%

Better than 56% of models compared

AIME

American Invitational Mathematics Examination

52.0%

Better than 72% of models compared

Math-500

Diverse mathematical problem solving benchmark

94.2%

Better than 76% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

83.7%

Better than 86% of models compared

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

75.1%

Better than 57% of models compared

Terminal-Bench Hard (legacy)

Agentic coding and terminal use

32.6%

Better than 76% of models compared

T²-Bench Telecom (legacy)

Conversational AI agents in dual-control scenarios

78.9%

Better than 69% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation