1,140 software engineering tasks with complex function calls across 139 libraries. Complete mode: code generation from docstring.
Large diverse task set; tests real library usage beyond simple algorithms; calibrated difficulty.
Some tasks require specific library versions; base models without instruction tuning may score lower.
Higher is better. Pass@1 percentage of tasks where generated code passes all test cases.
Max score: 100
Some tasks require specific library versions.
| # | Model | Score |
|---|---|---|
| 1 | 70.5 | |
| 2 | 67.6 | |
| 3 | 67.1 | |
| 4 | 65.8 | |
| 5 | 65.4 | |
| 6 | 65.3 | |
| 7 | 65.1 | |
| 8 | 64.7 | |
| 9 | 64.5 | |
| 10 | 64.3 | |
| 11 | 63.4 | |
| 12 | 63.4 | |
| 13 | 63.3 | |
| 14 | 63.1 | |
| 15 | 62.8 | |
| 16 | 62.7 | |
| 17 | 62.2 | |
| 18 | 62.1 | |
| 19 | 61.8 | |
| 20 | 61.7 | |
| 21 | 61.7 | |
| 22 | 61.5 | |
| 23 | 61.5 | |
| 24 | 61.4 | |
| 25 | 61.3 | |
| 26 | 61.2 | |
| 27 | 61.2 | |
| 28 | 61.2 | |
| 29 | 61.1 | |
| 30 | 61.0 | |
| 31 | 60.9 | |
| 32 | 60.8 | |
| 33 | 60.8 | |
| 34 | 60.8 | |
| 35 | 60.8 | |
| 36 | 60.7 | |
| 37 | 60.4 | |
| 38 | 60.3 | |
| 39 | 60.2 | |
| 40 | 60.2 | |
| 41 | 60.1 | |
| 42 | 60.1 | |
| 43 | 59.9 | |
| 44 | 59.9 | |
| 45 | 59.9 | |
| 46 | 59.8 | |
| 47 | 59.8 | |
| 48 | 59.7 | |
| 49 | 59.7 | |
| 50 | 59.6 |