Raschka deep dive: GPT-6 Astra, recurrent depth, and whether it hides its chain of thought
bibryam · x · 2026-09-09
Sebastian Raschka published a detailed article on GPT-6 Astra and the rumor that it 'hides' its reasoning trace, tying it to looped transformers and recurrent depth.
- Impressions: Astra is the best model he has used, leapfrogging GPT-5.6 across writing, math and coding, with especially strong 3D rendering/animation demos; benchmark results back this up
- He explains what looped transformers are (reusing transformer blocks) and how they relate to hidden chains of thought
- Surveys new insights from recent research papers on looping transformer blocks
A solid primer for anyone following the recurrent-depth architecture direction.
Related event: Raschka Breaks Down GPT-6 Astra and Looped Transformers(4 posts)→
More from Models
- OpenAI lists GPT-6.1 Terra: 1.05M context, $1/$5 pricing — but this poster sticks with Opus 5.5 — BLCNYY · 2026-10-27
- GPT-6.1 Terra cuts factual error rate from 22.3% to 4.3%, an ~81% drop — haider1 · 2026-10-27
- GPT-6.1 Terra lands: early users praise speed, value and subagent performance — BenBajarin · 2026-10-27
- Leaked GPT-6.1 Terra benchmarks: cheap but 'boring', still far behind Astra and Claude — Angaisb_ · 2026-10-27
- Artificial Analysis Ships Intelligence Index v4.3.2 With GPT-6.1 Terra Comparison — ArtificialAnlys · 2026-10-27
- GPT-6.1 Terra cuts hallucination rate by 50 points while lifting accuracy 2 points — ArtificialAnlys · 2026-10-27