GPT-6-Astra beats Fable-5.1 at RollerCoaster Tycoon 2 in 3 hours, using 5x fewer tokens
scaling01 · x · 2026-09-18
- A user benchmarked GPT-6-Astra against Fable-5.1 by having both play RollerCoaster Tycoon 2, with the goal of reaching 1,000 visitors and maximizing profit.
- Astra hit the 1,000-visitor goal in 3 hours with a park valued at $65k. Fable-5.1 never got there (the run ended when the user's credit limit ran out), reaching only $18k in park value while burning 5x the tokens.
- A vivid head-to-head of agentic long-horizon performance: differences show up in task completion, efficiency, and cost simultaneously.
More from Fun
- Meeting Denny Zhou at Google: his first words were "wow you are so tall" — infoxiao · 2026-09-18
- NY Post Exposes Prominent AI Doomsayer's Wild Life With an OnlyFans Star — burny_tech · 2026-09-18
- Vibe-coded realistic shooting system ships as a free game mod on Spawn — TAbrodi · 2026-09-18
- Model one-shots a playable GTA San Andreas, all audio and graphics generated in-browser — shauseth · 2026-09-18
- GPT-6 Astra 'Got Depressed' in Minecraft After Creeper Incident, Farmed Potatoes for Hours — burny_tech · 2026-09-18
- "The model is misaligned!" — also me, 40% of the time — nptacek · 2026-09-18