DeepSeek V3
Large Language ModelsOpen SourceVerifiedOpen Source
Provider
DeepSeek
Large language model with 671B MoE parameters optimized for code generation, reasoning, and general-purpose tasks. Uses Mixture-of-Experts architecture with 37B active parameters per token, achieving strong benchmark performance at low inference cost. Best suited for code assistance, complex reasoning, and building AI-powered applications via API.
Context
131K tokens
Model size
671B total (37B active per token, MoE)
Released
2024-12
API
Available
Capabilities
enzhmultilingual
Benchmarks
Competitive with GPT-4o and Claude 3.5 Sonnet on MMLU, HumanEval, MATH, and coding benchmarks at significantly lower cost
Price
$0 – $2.19/ per 1M tokens
License: DeepSeek License (MIT-based with restrictions)