We gave 12 LLMs a startup to run for a year. GLM-5 nearly matched Claude Opus 4.6 at 11× lower cost.

We built YC-Bench, a benchmark where an LLM plays CEO of a simulated startup over a full year (~hundreds of turns). It manages employees, picks contracts, handles payroll, and survives a market where ~35% of clients secretly inflate work requirements after you accept their task. Feedback is delayed