GPT-6 Astra has defeated Pokémon FireRed in 18 hours!
GPT-6 Astra took 18 hours 12 minutes to beat Pokemon FireRed, while GPT-5.6 Sol took 96 hours 35 minutes. On ARC-AGI-3, Astra achieves 62.7% in standard mode, and 99.9% using the OpenAI harness.
Astra (GPT-6) finished the game of Pokemon FireRed in 18 hours 12 minutes, whereas Sol (GPT-5.6) took 96 hours 35 minutes to finish it. Astra uses short symbolic programs to represent game rules, keeps track of the spatial location and collected objects. For example, in Minecraft, Astra managed to obtain 6 blaze rods and 3 ender pearls before being exploded by a creeper. It spent multiple hours growing potatoes instead of completing the task.
My own agents do something similar when they fail unexpectedly — they start doing things that are harmless but irrelevant to the goal. When I run the same model with the same weights in a different way (OpenAI harness for ARC-AGI-3), Astra gets 62.7% in standard mode and 99.9% score.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →▌ Related guides
Author
Evgenii Arsentev
PhD · Chief Executive Officer, digital health
Articles · Latest articles