DeepSeek Coder V2
DeepSeekCode Models
84.5
Performance
★ 4.5
Rating
445
Reviews
Text GenerationSmallCode AssistantOpen Weight
About
DeepSeek's specialized code model with strong performance on coding benchmarks at competitive pricing.
Strengths
Very strong at code generation (HumanEval) at lower cost than top generalists. MoE architecture for efficient inference. Good at multi-language code generation and debugging. Open-weight for self-hosting.
Specifications
- Context window
- 128,000
- Parameters
- —
Pricing
- Input cost
- $0.14 / 1M
- Output cost
- $0.28 / 1M
$0.14 input / $0.28 output per 1M tokens. Open-weight for self-hosting.
Speed & Latency
- 100
- tokens/sec
- 200ms
- time to first token
Available On
DeepSeek APIHuggingFaceTogether AI
Features
code completionfill in the middlestreamingmulti language
Performance Trend
Benchmark score trends over time for the top 5 benchmarks.
Loading history...
Benchmarks
Scores from various benchmark tests; higher is better.
| Test | Score | Percentile | Source |
|---|---|---|---|
| ARC-Challenge | 89.7 | — | huggingface |
| GSM8K | 83.2 | p87 | seed |
| GSM8K | 72.7 | — | huggingface |
| HumanEval |
