DeepSeekDeepSeek

DeepSeek V3

685B MoE model for general intelligence

DeepSeek V3 is a 685-billion parameter mixture-of-experts model that activates only 37B parameters per forward pass, making it highly efficient. It outperforms all previous open-source models on coding, math, and reasoning.

Try DeepSeek V3

Specifications

Context window163k tokens
Input price$0.270 / 1M tokens
Output price$1.10 / 1M tokens
Credits per query3 cr
Released2024-12
LicenseMIT

Features

Tool useJSON modeFunction callingCode

Benchmark Scores

What do these mean?

Sourced from official model cards and academic papers. Higher is better.

PhD-level science & reasoning

59.1%

Mathematical olympiad problems

Chatbot Arena human preference ranking

1277

Python coding — pass@1 rate

82%

Competition-level mathematics

89.1%

General knowledge across 57 academic domains

88.5%

Multi-turn instruction following (out of 10)

Real-world GitHub issue resolution rate