Neuro-Bench extends our bottom-up knowledge-graph reasoning evaluation to neuroscience. The suite comprises 5,000 high-quality neuroscience reasoning questions systematically generated from neuroscience knowledge-graph paths. Questions are organized by k-hop reasoning depth (1-Hop through 5-Hop), with 1,000 questions per hop for a balanced complexity distribution.
To keep the experience fresh, 250 questions are made available each week (50 from each hop), so the entire dataset is covered over a 20-week rotation. On week 21 the bins are reshuffled and the rotation repeats.
Single-step reasoning along one knowledge-graph relation
Two-step compositional reasoning across linked concepts
Mid-depth chains requiring multi-step integration
Deep relational chains spanning multiple subsystems
Long-range reasoning over five-step KG paths
1,000 per hop · 250 served each week · 20-week rotation
Each week’s session presents 250 questions — 50 from each k-hop bin, shuffled across hops so the difficulty mixes naturally.
Each item is a four-option clinical or experimental neuroscience question grounded in a KG path.
After each answer we visualise the underlying k-hop KG path along with a structured reasoning explanation.
A new 250-question set unlocks every Monday. The full 5,000-question dataset is exhausted in 20 weeks, then a reshuffled cycle begins.
Monitor your running score and per-hop performance as you work through the week’s items.
This week's set contains 250 questions — 50 from each of the 1- through 5-hop reasoning bins, shuffled across hops.