GPT-6 Astra One-Shots $10k Gambonanza Challenge in Obscure Game, Fully Autonomous
Vjeux · x · 2026-09-19
Frustrated by claims that LLMs are AGI when they can't play random Steam games, Vjeux set up a $10k Gambonanza challenge on an obscure game almost certainly absent from training data.
GPT-6 Astra won on its first attempt: 4 hours, fully autonomous, using only screenshots plus mouse/keyboard inputs with no game memory access. The runner noted its standout vision and spatial reasoning — other vision models couldn't even track pieces on the board, while Astra read the state correctly and planned a winning build.
For comparison, Claude Fable 5.1 was still stuck on the second boss after 4 hours. Astra remains the only model to finish the game, and the prompt used was fairly basic.
More from Models
- ChatGPT Pro 20x plan back on sale after OpenAI's capacity pause — mark_k · 2026-09-19
- OrukLabs launches Resonance-2 with 31 emotion and speaking-style signals from audio — ChrisGPotts · 2026-09-19
- OpenAI launches Astra for Law; lawyer slams ZDR privacy promises as unverifiable — bgmshana · 2026-09-19
- Noam's Actual Quote: Multi-agent Credited Less Than 10% for Math Breakthrough — eliebakouch · 2026-09-19
- LLM logprobs as soft classifiers: a year-old idea finally validated — JnBrymn · 2026-09-19
- Jev benchmarks: matches production classifiers on fixed-label tasks at ~100x lower cost — ivan_bezdomny · 2026-09-19