Overview
- An experimenter known as CozyBlaze reported that GPT-6 Astra autonomously completed Valve’s 3D puzzle game Portal by controlling the game through a Model Context Protocol and a modified SourcePauseTool.
- The full playthrough was recorded across about 24 hours because the system frequently paused for the model to plan moves, and the footage was edited down to roughly two hours of highlights.
- CozyBlaze logged 3,336 tool calls for the run and estimated an API cost of $571.18, which they said was covered by a $200-per-month Codex Pro subscription.
- OpenAI presents GPT-6 Astra as its new flagship model with state-of-the-art performance claims across technical and professional tasks, and the company indicated more demonstrations are expected soon.
- CozyBlaze emphasized the result is a pragmatic demonstration rather than an independent benchmark, and the run has not been independently verified or broadly benchmarked.