Act Wisely: Cultivating Meta-Cognitive Tool Use in Agentic Multimodal Models

Open paper

Benchmark	Model	Rank	Results
chart-question-answering-on-charxiv-dq	Metis	#10	Accuracy: 83.4
chart-question-answering-on-charxiv-rq	Metis	#21	Accuracy: 54.1
multimodal-reasoning-on-mathvista	Metis	#36	Accuracy: 78.0