Collaborator review

ZooBench campaign reviews

Subject-level model results, benchmark targets, and volunteer context for ten citizen-science campaigns.

Current review set

Choose a campaign

10 campaigns · 100 subjects each

01Science Scribbler: Correct MitoGPT-6 Astra · low reasoning83.6%02Etch A Cell: Going NuclearGPT-6 Astra · low reasoning76.5%03Backyard Worlds: Cool NeighborsGPT-6 Astra · low reasoning87.0%04Rubin Alert ExplorersGPT-6 Astra · low reasoning94.0%05Gravity SpyGPT-6 Astra · low reasoning50.0%06Infection InspectionGPT-6 Astra · low reasoning71.0%07Synaptic Protein ZooGPT-6 Astra · low reasoning92.5%08Dark Energy ExplorersGPT-6 Astra · low reasoning80.0%09Galaxy Zoo: Tidal TalesGPT-6 Astra · low reasoning63.6%10Penguin WatchGPT-6 Astra · low reasoning25.5%

Each page shows one complete GPT-6 Astra direct API run at low reasoning, its tutorial, and all 100 subject results. Scores use each campaign’s current metric and reference labels; a percentage represents mean task score. Updated October 22, 2026.