SLDBench: leaderboard
Scaling Law Discovery benchmark evaluating AI agents on discovering mathematical scaling laws from experimental LLM training data.
Metric: Mean Reward R2 (score). Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Around April 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 0.48 | #66 |
| 2 | Claude Sonnet 4.5 | 0.42 | #142 |
| 3 | GPT-5 | 0.34 | #98 |
| 4 | Claude Haiku 4.5 | 0.22 | #280 |
| 5 | GPT-5.2 | 0.12 | #110 |
| 6 | GPT-4.1 | 0.11 | #248 |
| 7 | Gemini 3 Flash (Preview) | -0.03 | #82 |
| 8 | O4 Mini | -0.26 | #180 |
| 9 | O3 | -0.27 | #124 |
| 10 | DeepSeek V3.2 (Thinking) | -0.5 | #205 (DeepSeek V3.2) |
| 11 | DeepSeek V3.2 | -0.56 | #205 |
| 12 | Gemini 2.5 Flash | -0.75 | #245 |
| 13 | GPT-4o | -1 | #343 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sldbench · How It Works · Data refreshed daily, snapshot 2026-10-20.