Astra's perception got much better: mostly scale plus pretraining encoder wins
willdepue · x · 2026-09-07
In the ongoing discussion on Astra's 3D modeling leap, willdepue notes the model's perception improved dramatically — "probably mostly just scale, but I'm sure some pretraining arch/encoder win." Astra "can see really well and this surely helps." Combined with his earlier take, he attributes the overall jump largely to better-targeted RL environments.
Related event: Debate Over Astra's 3D Leap: No-CoT Ability or Scale?(3 posts)→
More from Models
- Google's Astra Ultra splits games vs 1800-rated chess bot, resigns after missing pawn promotion — MikePFrank · 2026-09-07
- Speculation: ChatGPT's model may be a twice-distilled version of the 100K B200 run — teortaxesTex · 2026-09-07
- Dev impressed by Gemini 3.8 Flash and Antigravity's clever Unicode math formatting — doodlestein · 2026-09-07
- ML researcher pushes back: Astra 'has been disappointing' beyond Blender demos — gowthami_s · 2026-09-07
- Developer: Grok and GPT-6 are scarily good at reverse-engineering desktop apps — jasonkneen · 2026-09-07
- "May be remembered as the moment true AGI arrived": GPT-6 Astra post hits 125M views — DeryaTR_ · 2026-09-07