LiveBench's instruction following category — measures how precisely models follow complex, multi-constraint instructions.
Tests whether models follow explicit formatting and content constraints. Useful for applications requiring structured output.
Benchmark instructions are cleaner than real user requests, so ambiguity handling is under-tested; high compliance scores do not guarantee downstream factual reliability.
Higher is better. Accuracy of instruction adherence.
Max score: 100
Benchmark instructions are cleaner than real user requests, so ambiguity handling is under-tested.
| # | Model | Score |
|---|---|---|
| 1 | 81.4 | |
| 2 | 79.9 | |
| 3 | 79.1 | |
| 4 | 78.0 | |
| 5 | 78.0 | |
| 6 | 77.1 | |
| 7 | 75.8 | |
| 8 | 75.6 | |
| 9 | 75.6 | |
| 10 | 75.4 | |
| 11 | 75.3 | |
| 12 | 74.3 | |
| 13 | 74.2 | |
| 14 | 74.1 | |
| 15 | 74.0 | |
| 16 | 73.4 | |
| 17 | 72.7 | |
| 18 | 72.0 | |
| 19 | 71.9 | |
| 20 | 71.9 | |
| 21 | 71.8 | |
| 22 | 71.5 | |
| 23 | 71.5 | |
| 24 | 71.4 | |
| 25 | 71.0 | |
| 26 | 70.7 | |
| 27 | 70.5 | |
| 28 | 70.2 | |
| 29 | 70.1 | |
| 30 | 70.0 | |
| 31 | 69.6 | |
| 32 | 69.3 | |
| 33 | 68.6 | |
| 34 | 67.7 | |
| 35 | 67.7 | |
| 36 | 67.5 | |
| 37 | 67.2 | |
| 38 | 67.2 | |
| 39 | 66.5 | |
| 40 | 66.4 | |
| 41 | 65.5 | |
| 42 | 65.2 | |
| 43 | 64.6 | |
| 44 | 64.4 | |
| 45 | 63.9 | |
| 46 | 63.8 | |
| 47 | 63.1 | |
| 48 | 62.8 | |
| 49 | 62.4 | |
| 50 | 62.3 |