Post-Cutoff

Astra’s no-chain-of-thought capability jump replicates

Neel Nanda @NeelNanda5XImportance: major (4 of 5)

Why it matters

An independent replication by DeepMind’s interpretability lead supporting the claim that Astra does far more computation without verbalized reasoning, which is the core of the monitorability debate.

Summary

Neel Nanda (Google DeepMind mechanistic interpretability lead) wrote that the Astra system card’s claim that the model can do a lot of computation without chain of thought “replicates”. In his test Astra managed about 1.75x the steps of the next-best models (Fable 5.1, Gemini 3.8 Flash) without CoT. He noted that no-CoT capabilities had risen much faster than with-CoT capabilities, calling it “a concerning trend”. The data supports the argument (Zvi Mowshowitz, Rob Wiblin, Gary Marcus) that the more models can compute per forward pass, the less they need to verbalize, which erodes CoT monitoring. Nanda co-authored the July 2025 “Chain of Thought Monitorability” position paper. Verified via the X syndication API (2026-09-10 22:32 UTC, ~1.3K likes).

Archived text

The Astra system card claims it can do a lot of computation without chain of thought

This replicates: Astra is a massive jump, doing 1.75x the steps of the next best models (Fable 5.1/Gemini 3.8 Flash)

No CoT capabilities went up far more than those with CoT, a concerning trend

Source: x.com/NeelNanda5/status/2098177895932068174Archived 2026-09-29 via syndication.Counts: 1,256 likes, 42 replies (at fetch time)Media: image

Cited in

  1. Model releases 65 days after the cutoff

    OpenAI releases GPT-6 Astra, its first GPT-6 model

People in this post

Gary Marcus, Cognitive scientist and author; professor emeritus, NYU; Zvi Mowshowitz, Writer, 'Don't Worry About the Vase'