ResearchBenchmarksStir on scientific experiment datasetFollow74.4Success RateCAPER28.43240.36652.364.234Feb 10, 2026Evaluation ResultsMethodMethodLinksSuccess RateCAPER2026.0274.4MP+MLM+RLBackbone=GPT-4oBackbone=GPT-4o2026.0270.8MLM+RLBackbone=GPT-4oBackbone=GPT-4o2026.0270.5MLM+RLBackbone=LLaMA3.2, Rea...Backbone=LLaMA3.2, Reasoning Strategy=w/o CoT2026.0239.6LLM+RLBackbone=GPT-4oBackbone=GPT-4o2026.0234.8LLM+RLBackbone=Llama-3.1-8B,...Backbone=Llama-3.1-8B, Reasoning Strategy=w/o CoT2026.0230.2