SLR-Bench - Basic: leaderboard
The Basic tier of SLR-Bench (AIML lab, TU Darmstadt) — systematically generated logical-reasoning puzzles where models must induce first-order logic rules that separate positive from negative examples, verified symbolically; accuracy (%).
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 56 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 100 | #152 |
| 2 | GPT-5 | 100 | #98 |
| 3 | Gemini 2.5 Flash | 100 | #245 |
| 4 | Qwen 3.5 9B | 100 | #379 |
| 5 | Qwen 3.5 27B | 100 | #187 |
| 6 | Gemini 3 Flash (Preview) | 100 | #82 |
| 7 | GPT-OSS-120B (High) | 100 | #341 (GPT-OSS-120B) |
| 8 | O3 Mini | 100 | #272 |
| 9 | GPT-5 Mini (High) | 100 | #186 (GPT-5 Mini) |
| 10 | GPT-5 Chat | 100 | #241 |
| 11 | GPT-OSS-120B (Medium) | 100 | #341 (GPT-OSS-120B) |
| 12 | GPT-5 Nano (High) | 100 | #425 (GPT-5 Nano) |
| 13 | GPT-5 Mini (Medium) | 100 | #186 (GPT-5 Mini) |
| 14 | GPT-5 Mini (Low) | 100 | #186 (GPT-5 Mini) |
| 15 | Gemini 3.1 Flash Lite (Medium) | 100 | #191 (Gemini 3.1 Flash Lite) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=slr-bench-basic · How It Works · Data refreshed daily, snapshot 2026-10-20.