Magicoder-S-DS-6.7B: benchmark results
ise-uiuc's Magicoder-S-DS 6.7B, fine-tuned from DeepSeek-Coder 6.7B on OSS-Instruct and Evol-Instruct coding datasets generated with GPT-3.5-turbo. Provider: UIUC. Released 2023-12-03. Access: Open.
Aggregate Rating 1391 ± 24, rank #1232 of 1640 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EvoEval Subtle | 67 | Pass@1 (%) | 90 |
| EvoEval Difficult | 40 | Pass@1 (%) | 88 |
| EvoEval | 54.48 | Pass@1 (%) | 86 |
| EvalPlus (HumanEval+ & MBPP+) | 70.2 | Pass@1 avg (%) | 81 |
| EvoEval Tool Use | 61 | Pass@1 (%) | 81 |
| EvoEval Verbose | 69.51 | Pass@1 (%) | 80 |
| InfiBench | 51.46 | Score (%) | 80 |
| EvoEval Combine | 21 | Pass@1 (%) | 79 |
| McEval | 48.6 | AVG (%) | 76.2 |
| CodeEditorBench Plus - Code Debug (Three-shot) | 26.2 | pass@1 (%, greedy decoding, three-shot, CodeEditorBench_Plus | 72.2 |
| CodeEditorBench Plus - Code Requirement Switch (Three-shot) | 19.2 | pass@1 (%, greedy decoding, three-shot, CodeEditorBench_Plus | 72.2 |
| CodeEditorBench Plus - Code Requirement Switch (Zero-shot) | 13 | pass@1 (%, greedy decoding, zero-shot, CodeEditorBench_Plus) | 72.2 |
Interactive version: theaggregate.ai/model?slug=magicoder-s-ds-6-7b · How It Works · Data refreshed daily, snapshot 2026-11-07.