DeepSeek V3

Large Language ModelsOpen SourceVerifiedOpen Source

Provider

DeepSeek

Chinese open-weight model rivaling GPT-4o. MoE architecture, 671B params. Exceptional value for reasoning and coding.

Context

128K tokens

Model size

671B (37B active)

Released

2024-12

API

Available

Capabilities

textmultilingual

Benchmarks

mmlu88.5%
humaneval89.9%

Price

From $0

License: MIT