deepseek-coder-6.7B-instruct: benchmark results
DeepSeek's 6.7B instruct code model, initialized from deepseek-coder-6.7b-base and pretrained on 2 trillion tokens (87% code, 13% English/Chinese text) with a 16K context window. Provider: DeepSeek. Released 2023-10-29. Access: Open.
Aggregate Rating 1370 ± 17, rank #1335 of 1656 rated models, from 42 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Big Code Memorization - HumanEval-ET pass@1 | 62.8 | HumanEval-ET pass@1 (%) | 100 |
| EvoEval Verbose | 72.56 | Pass@1 (%) | 91 |
| EvoEval Concise | 71.95 | Pass@1 (%) | 89 |
| Big Code Memorization - HumanEval pass@1 | 71.34 | HumanEval pass@1 (%) | 88.2 |
| Big Code Memorization - HumanEval pass@50 | 64.94 | HumanEval pass@50 (%) | 88.2 |
| Big Code Memorization - HumanEval-ET pass@50 | 56.73 | HumanEval-ET pass@50 (%) | 88.2 |
| EvoEval Difficult | 40 | Pass@1 (%) | 88 |
| EvoEval | 53.24 | Pass@1 (%) | 80 |
| EffiBench - NMU | 1.9 | Normalized Memory Usage | 79.3 |
| EvalPlus (HumanEval+ & MBPP+) | 68.4 | Pass@1 avg (%) | 78.2 |
| EffiBench - NET | 2.89 | Normalized Execution Time | 74.4 |
| EvoEval Creative | 37 | Pass@1 (%) | 74 |
Interactive version: theaggregate.ai/model?slug=deepseek-coder-6-7b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.