agents benchmark
SkillsBench evaluates coding agents on self-contained programming tasks, measuring practical engineering skills across diverse software development scenarios.
Updated Aug 7, 2026
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 70.2% | 100.0% | 7 | C | |
| 2 | AC | 59.2% | 83.3% | 7 | C | |
| 3 | TE | 55.3% | 66.7% | 7 | C | |
| 4 | AC | 54.9% | 50.0% | 7 | C | |
| 5 | AC | 48.2% | 33.3% | 7 | C | |
| 6 | AC | 45.7% | 16.7% | 7 | C | |
| 7 | AC | 28.7% | 0.0% | 7 | C |
Top published rows on the benchmark's original scale.
The top published results on this benchmark's own scale.
Definition and scoring fields from the benchmark registry.
SkillsBench evaluates coding agents on self-contained programming tasks, measuring practical engineering skills across diverse software development scenarios.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
Common questions about SkillsBench.
Qwen3.8 Max is currently ranked first with 70.2%.
SkillsBench evaluates coding agents on self-contained programming tasks, measuring practical engineering skills across diverse software development scenarios.
Yes. Higher values rank better for this benchmark.
7 unique published model results are currently shown.
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.