SDABench - Mechanistic: leaderboard
Metric: Accuracy (%) of open-ended answers on the held-out SDA-Synth test split (the 400 mechanistic analysis instances; zero-shot, JSON reasoning and answer fields, the answer scored by relative-error numeric, exact or structured matching with a GPT-4o fallback judge); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Kimi K2.5 | 70 |
| 2 | Qwen 3.5 397B A17B | 48 |
| 3 | GPT-5.4 | 47 |
| 4 | Claude Sonnet 4.6 | 45 |
| 5 | DeepSeek V3.2 | 41 |
| 6 | Gemini 3.1 Pro (Preview) | 35.75 |
| 7 | GLM-5 | 30.25 |
| 8 | DeepSeek R1 | 23.25 |
| 9 | Qwen 3 235B A22B 2507 Instruct | 22.5 |
| 10 | GPT-5.4 Mini | 18.75 |
| 11 | Llama 3.3 70B Instruct | 2.75 |
| 12 | Llama 3.1 8B Instruct | 0 |
Interactive version: theaggregate.ai/benchmark?slug=sdabench-mechanistic · How It Works · Data refreshed daily, snapshot 2026-09-29.