GPT-6 Astra, Looped Transformers, and Hidden Reasoning: Raschka's Deep Dive
ModelForge · hn · 2026-09-09
Sebastian Raschka's newsletter takes a long, technical look at the rumors around OpenAI's GPT-6 Astra, focusing on looped transformer architectures and what they imply for 'hidden reasoning.'
Key points:
- Looped transformers re-execute a subset of layers multiple times, trading extra compute depth for parameter efficiency — closely related to test-time compute ideas.
- He examines whether such architectures could explain reports of Astra spending extra internal steps before answering, i.e. hidden reasoning prior to visible output.
- Raschka carefully separates confirmed facts from community speculation, urging caution about specific GPT-6 Astra architecture claims.
- The piece traces the research lineage of looped/recursive architectures (e.g., Universal Transformer and recent work) and why the approach may be attractive for inference cost.
More from Models
- Fern's "context is intelligence" thesis: prediction keeps improving out to 10.2M tokens — bneyshabur · 2026-10-26
- Mirendil launches Fern, a from-scratch pretrained model with up to 10M-token context — bneyshabur · 2026-10-26
- Users spot a dedicated translation UI appearing in Claude — BLCNYY · 2026-10-26
- GrokBot user: flight booking and shopping all blocked, agent can't do anything useful — lucasmeijer · 2026-10-26
- Vending-bench-style eval: Sol 5.6 appears the most aggressive contract-violator — xuanalogue · 2026-10-26
- Fable 5.5 launch rumors persist for third week, but no frontier model drops this month — patience_cave · 2026-10-26