DeepSeek-Coder-V2
Open Weights
DeepSeek
DeepSeek's open-source code MoE (June 2024), further pretrained from DeepSeek-V2 on an additional 6T tokens. It expands programming-language support from 86 to 338 and context from 16K to 128K, reaching performance comparable to GPT-4 Turbo on code-specific tasks. Available in 236B (21B active) and 16B (2.4B active) sizes.
Strengths
- Open-source code MoE rivaling GPT-4 Turbo on coding tasks
- Supports 338 programming languages (up from 86)
- 128K context for repository-scale work
- Efficient MoE: 236B total with 21B active (also a 16B Lite variant)
- Commercial use permitted under the DeepSeek License
Caveats
- Text-only, specialized for code over general chat
- Superseded by newer DeepSeek-V3 generation for general tasks
- Full 236B model demands substantial GPU memory
- Vendor
- DeepSeek
- Context Window
- 128,000 tokens
- License
- DeepSeek License
- Release Date
- 2024-06-17
- Modalities
- text
| Benchmark | Score | Max |
|---|---|---|
| livebench_language | 33.0 | 100.0 |
| livebench_coding | 41.5 | 100.0 |
| livebench_instruction_following | 75.8 | 100.0 |
Capabilities
Vision
Audio
Video
Tool Use
Pricing
Open weights under the DeepSeek License (commercial use permitted) - free to self-host; hosted API providers vary.