LiveBench's coding category — evaluates code generation with new problems from recent programming contests.
Fresh contest problems reduce memorization advantage from public training data; execution-focused scoring rewards syntactic and semantic correctness.
Contest-style tasks overweight algorithmic puzzle solving relative to production maintenance work; difficulty shifts across monthly updates can make longitudinal comparisons noisy.
Higher is better. Pass rate on fresh coding problems.
Max score: 100
Contest-style tasks overweight algorithmic puzzle solving relative to production maintenance work.
| # | Model | Score |
|---|---|---|
| 1 | 98.3 | |
| 2 | 96.0 | |
| 3 | 94.4 | |
| 4 | 94.3 | |
| 5 | 93.1 | |
| 6 | 91.9 | |
| 7 | 91.4 | |
| 8 | 91.0 | |
| 9 | 90.8 | |
| 10 | 90.2 | |
| 11 | 89.7 | |
| 12 | 89.7 | |
| 13 | 89.7 | |
| 14 | 89.2 | |
| 15 | 87.5 | |
| 16 | 87.4 | |
| 17 | 87.1 | |
| 18 | 85.7 | |
| 19 | 85.1 | |
| 20 | 84.4 | |
| 21 | 84.2 | |
| 22 | 84.0 | |
| 23 | 83.9 | |
| 24 | 83.9 | |
| 25 | 83.7 | |
| 26 | 83.6 | |
| 27 | 83.3 | |
| 28 | 83.2 | |
| 29 | 82.8 | |
| 30 | 82.7 | |
| 31 | 82.5 | |
| 32 | 82.4 | |
| 33 | 82.1 | |
| 34 | 82.1 | |
| 35 | 82.1 | |
| 36 | 81.3 | |
| 37 | 81.2 | |
| 38 | 81.2 | |
| 39 | 81.1 | |
| 40 | 80.7 | |
| 41 | 80.7 | |
| 42 | 80.7 | |
| 43 | 80.6 | |
| 44 | 80.5 | |
| 45 | 80.4 | |
| 46 | 80.4 | |
| 47 | 79.9 | |
| 48 | 79.8 | |
| 49 | 79.8 | |
| 50 | 79.8 |