Portal2Bench tests whether AI agents can solve Portal 2 chambers

Portal2Bench, an independent benchmark not affiliated with Valve, measures whether AI agents can solve 26 Portal 2 test chambers from chapters 2 to 5, scoring a chamber solved only if the agent reaches the next map within 300 plans and $30. GPT-6.1 Sol and GPT-6 Astra, both run through Codex, each solved 20 of 26 chambers, GPT-6.1 Terra solved 19, Opus 5.5 solved 11, Sonnet 5.5 solved 6, and GPT-6 Luna solved 0. The agent sends key presses and view turns and receives screenshots and its position, with access only to a notes folder and no shell, web or console; median cost per solved chamber ranges from $0.74 for GPT-6.1 Terra to $13.12 for Fable 5.1.