Zhejiang University's BrainBench covers 4 subsets, 17 datasets, and 3,000+ EEG tasks on real data?

Zhejiang University's BrainBench covers 4 subsets, 17 datasets, and 3,000+ EEG tasks on real data

LLMs evaluated across 100K+ executions using CodeAct and BrainAgent agentic paradigms?

LLMs evaluated across 100K+ executions using CodeAct and BrainAgent agentic paradigms

Outputs validated via numerical, categorical, set, sequence, semantic, and artifact checks

Research & Papers

arXiv cs.AI August 06, 2026

⚡From sleep scoring to neurocognition, this benchmark evaluates LLMs on 3,000+ real EEG tasks

Deep Dive

Key Points

Zhejiang University's BrainBench covers 4 subsets, 17 datasets, and 3,000+ EEG tasks on real data
LLMs evaluated across 100K+ executions using CodeAct and BrainAgent agentic paradigms
Outputs validated via numerical, categorical, set, sequence, semantic, and artifact checks

As LLMs enter neuroscience, BrainBench sets a standard for measuring real-world EEG analysis capability.