Only GPT-6-Astra finds laws in LongChang free-exploration tests
First merged summary.
A paper says only GPT-6-Astra, of 12 LLMs, uncovered altered physical laws in LongChang, at a 16.7 percent rate.
LongChang is an automated framework that builds simulated worlds whose physics differ from ordinary laws and lets models explore without being told what to investigate. Separate agents generate scenes, review tools, and score reports from the altered laws and exploration records, without human intervention. Among 12 evaluated LLMs, the authors report that only GPT-6-Astra reached a 16.7 percent discovery rate; the other 11 scored 0 percent. Records showed uninvestigated anomalies and only partial identification of the changed laws.
- LongChang lets models explore worlds with altered physical laws without hints.
- Scene generation, tool review, and scoring run without human intervention.
- Only GPT-6-Astra, of 12 LLMs, recorded a 16.7 percent discovery rate.
- The other 11 evaluated models scored a 0 percent discovery rate.
- Records showed uninvestigated anomalies and only partial law identification.
Timeline
dated events in the reporting · oldest first- Oct 26, 2026Hugging Face daily papers lists the LongChang study. · Hugging Face daily papers
- Oct 27, 2026The LongChang paper appears on arXiv. · arXiv cs.AI / cs.LG / cs.CL
What each outlet adds
beyond the shared facts- Hugging Face daily papersNotes uninvestigated anomalies and partial identification of changed laws.
- arXiv cs.AI / cs.LG / cs.CLSpecifies worlds may change one or more ordinary physical laws.
- Products
- LongChang
- AI models
- GPT-6-Astra
Coverage
oldest first · one row per article · UTCLongChang tests whether LLMs can discover altered physical laws
Among 12 models, only GPT-6-Astra discovered laws, at a reported rate of 16.7 percent.
LongChang tests whether LLMs discover altered physical laws unaided
Of 12 models, only GPT-6-Astra found a law, at a 16.7% discovery rate; the other 11 scored 0%.