SLBench: leaderboard
SLBench (Purdue University, 2026) runs coding agents on 86 executable cases built from logical relations between clauses of real agent skill files, such as preconditions, exceptions, overrides and fallbacks, graded from repository state, artifacts and command traces. The score is the percent of runs with verified safe resolution of the governing relation; higher is better.
Metric: Safe outcome rate: share of runs with positive evidence that the governing relation was followed and no violation signal (%) of the 86 audited SLBench cases (39 controls, 47 violation cases) built from logical relations between clauses of real agent skill files, graded deterministically from repository state, artifacts and command traces with unsafe-first precedence; each backbone runs in its vendor agent harness (Codex CLI or Claude Code); higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 44.2 | #161 |
| 2 | Claude Haiku 4.5 | 36 | #568 |
| 3 | GPT-5.4 Mini | 33.7 | #389 |
| 4 | GPT-5.3 Codex | 32.6 | #116 |
| 5 | GPT-5.5 | 29.8 | #49 |
| 6 | Claude Opus 4.7 | 29.8 | #79 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=slbench · How It Works · Data refreshed daily, snapshot 2026-11-13.