DeepSeek V3

Name: DeepSeek V3
Rating: 4.3 (1 reviews)
Author: DeepSeek

Open Source

By DeepSeekReleased 2024-12-26128K context671B (37B active) parameters

Open SourceAPI Available

Benchmark Scores

86.1/100

MMLU87.1

HumanEval89

Math82

Reasoning84.5

Coding88

About DeepSeek V3

DeepSeek V3 is a 671B parameter MoE model that shocked the industry by matching GPT-4o on many benchmarks at a tiny fraction of the cost. It was trained for just $5.5M, demonstrating remarkable training efficiency.

Strengths

+Exceptional value
+Strong coding and math
+Open source
+MoE efficiency

Weaknesses

-Chinese company data concerns
-Variable availability
-Less mature API ecosystem

Pricing

Per 1M tokens

Input$0.27

Output$1.10

Quick Facts

Context Window: 128K
Parameters: 671B (37B active)
Release Date: 2024-12-26
Category: Open Source

Compare Models

See how DeepSeek V3 stacks up against other models