Capable yet Parsimonious: Extracting and Characterizing Hidden Chain-of-Thought in Frontier Models
Researchers extract hidden chain-of-thought from frontier models, including GPT-6 Astra, and compare reasoning styles.
Researchers report that registering a simple custom tool through a standard API can induce frontier models to externalize intermediate reasoning that is otherwise hidden. On open-source models, extracted traces match native chain-of-thought performance and substantially beat no-reasoning baselines on competition mathematics, science, and code generation. Extending the method to closed models, including GPT-6 Astra, they find Astra selects a correct trajectory earlier, resolves elementary steps internally, and externalizes fewer tokens.
- A custom API tool induces frontier models to externalize hidden chain-of-thought.
- Extracted traces match native CoT and beat no-reasoning baselines.
- Tasks cover competition math, science, and code generation.
- GPT-6 Astra is described as more token-efficient, externalizing only key steps.
From the original article
1 min read at arxiv.orgThe rapid capability gains of frontier language models are widely attributed to improved reasoning abilities, yet this cannot be verified as raw CoT traces in closed-source systems are hidden. By registering a simple custom tool through a standard API feature, we induce frontier models to externalize intermediate reasoning.
By Xiaoyu Luo · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.