OfficeTown Benchmark: GPT-6 Astra Tops at 74.4% on 406 Simulated Workplace Tasks With Persistent State

OfficeTown simulates 23 professional organizations across 139 job titles with persistent, permissioned workspaces where artifacts and decisions accumulate across tasks. On its 406-task bench (4,836 criteria from 71 domain experts), GPT-6 Astra leads at 74.4% Pass@1, Grok-4.7 at 58.4%, Claude Opus 5.5 at 50.7%, GPT-5.6 Terra at 39.4% — and even the top model misses P0 criteria on 20.8% of tasks. Fine-tuning Qwen3.5-9B on OfficeTown data lifted GDPval score +7.90 points.