Model
Inference & ServingCoding
DeepSeek-Coder-V2
1Open 2MoE 3model 4matching 5GPT-4 6coding
the six Ws · specification
W1
Who
DeepSeek AI developed DeepSeek-Coder-V2 as a successor to the original DeepSeek-Coder.
W2
What
A mixture-of-experts code and math model supporting 338 programming languages and long context.
W3
Where
Released on Hugging Face and GitHub under the deepseek-ai organization.
W4
When
DeepSeek-Coder-V2 was released in June 2024.
W5
Why
It claimed performance comparable to GPT-4 Turbo on coding and math benchmarks while remaining open weight.
W6
With
Runs via vLLM, SGLang, or transformers, using a 236B-parameter MoE architecture with 21B active parameters.
W7
Watch
Open-weight release by DeepSeek AI under the DeepSeek license, permitting commercial use with some conditions.
alternatives
works with