ResearchThe Decoder
GPT-6 Astra appears to show a "step change" in spatial reasoning based on early benchmarks
Summary
In a new robotics benchmark, GPT-6 Astra shows major gains in spatial understanding. On StationeryBench, the model completed 7 out of 100 tasks with dual-arm robots, while competitor MolmoAct2 couldn't finish a single one.
Original Article
Captured source content or English translation, normalized into this reading format.
This story does not yet have captured source text. Open the source link to read it.
Region
Europe
Heat Score
89
Category
Research
Language
en