SDABench - Predictive (MCQ): leaderboard

Metric: Accuracy (%) of four-option multiple-choice answers with perturbation-based distractors on the held-out SDA-Synth test split (the 400 predictive analysis instances; zero-shot; 25% chance); higher is better. Source: arxiv.org. Saturation forecast: Around December 2027. 13 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)63.75
2GPT-5.458.25
3Qwen 3 235B A22B 2507 Instruct49.5
4Claude Sonnet 4.648
5DeepSeek V3.247.75
6GLM-541
7GPT-5.4 Mini39.75
8DeepSeek R136.5
9Qwen 3.5 397B A17B35.25
10Llama 3.3 70B Instruct28.25
11Kimi K2.525.75
12Llama 3.1 8B Instruct24.75

Interactive version: theaggregate.ai/benchmark?slug=sdabench-predictive-mcq · How It Works · Data refreshed daily, snapshot 2026-09-29.