LDU-Bench - Cause Analysis: leaderboard
Image-conditioned cause attribution on 532 expert-reviewed lithography samples in LDU-Bench (2026). Deterministic rubric score combining cause-label match and rationale keyword overlap, in percent; higher is better.
Metric: Rubric score (%; 0.7 x cause-label match, 1 for an exact or alias match and 0.5 within the same supergroup, plus 0.3 x keyword F1 of the short rationale, on 532 expert-reviewed samples; IC-SEM lithography and integrated-circuit review images; one zero-shot run per model at temperature 0 where exposed; invalid outputs stay in the denominator; deterministic scoring, no LLM judge). Source: arxiv.org. Saturation forecast: Around March 2028. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 40.9 | #121 |
| 2 | MiniMax-M3 | 39.5 | #265 |
| 3 | Qwen 3.6 Plus | 37.6 | #217 |
| 4 | GLM-5V Turbo | 29.7 | #342 |
| 5 | Claude Opus 4.6 | 25.5 | #100 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ldu-bench-cause-analysis · How It Works · Data refreshed daily, snapshot 2026-11-12.