Research
Here's a link to the article: https://magazine.sebastianraschka.com/p/gpt-6-astra-looped-transformers-and
Sebastian Raschka’s September 2026 article introduces **GPT‑6 *Astra***, a next‑generation language model that expands upon the Transformer backbone with a novel *looped* architecture—allowing multipl
Sebastian Raschka’s September 2026 article introduces GPT‑6 Astra, a next‑generation language model that expands upon the Transformer backbone with a novel looped architecture—allowing multiple inference passes through the same attention layers to refine outputs iteratively. The paper details how these looped transformers enable “hidden reasoning”, wherein the model internally generates intermediate logical steps before producing a final answer, improving transparency and reducing hallucinations compared to earlier GPT versions. Raschka evaluates Astra on standard benchmarks, reporting state‑of‑the‑art performance and reduced parameter counts thanks to efficient reuse of transformer blocks.
Related
- A lot of hype around OpenAI's Astra model here on my timeline today. Apparently, this goes back to a new article from The Information, which…
- I put together a mega write-up on GPT-6 Astra & looped transformers. How looped transformers / recurrent depth works, cost-tradeoffs, whethe…
- GPT-6 Astra, Looped Transformers, and Hidden Reasoning
Source: Sebastian Raschka (X) | 2026-09-09