Research
GPT-6 Astra, Looped Transformers, and Hidden Reasoning
GPT‑6 Astra is OpenAI’s latest flagship model, outperforming GPT‑5.6 across virtually all common benchmarks—most noticeably in math, coding, and especially 3‑D rendering/animation tasks. The architect
GPT‑6 Astra is OpenAI’s latest flagship model, outperforming GPT‑5.6 across virtually all common benchmarks—most noticeably in math, coding, and especially 3‑D rendering/animation tasks. The architecture incorporates looped transformer blocks (a form of recurrent depth) that researchers speculate may hide the chain‑of‑thought reasoning trace used during inference. Recent papers on looped transformers are cited to explore how this looping design influences both model capacity and the handling of hidden reasoning chains.
Related
- I put together a mega write-up on GPT-6 Astra & looped transformers. How looped transformers / recurrent depth works, cost-tradeoffs, whethe…
- Here's a link to the article: https://magazine.sebastianraschka.com/p/gpt-6-astra-looped-transformers-and
- Do Transformers Use their Depth Adaptively? Evidence from a Relational Reasoning Task
- Asking the Right Questions: Improving Reasoning with Generated Stepping Stones
Source: Sebastian Raschka | 2026-09-09