# The Detection-Extraction Gap: Models Know the Answer Before They Can Say It

Source: [The Detection-Extraction Gap: Models Know the Answer Before They Can Say It](<https://www.sota2.com/research/paper/69d7327f7d0172140ec76af9>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-11T05:00:11.573Z

Authors: Hanyang Wang, Mingxuan Zhu

Published: 2026-04-08T02:47:31.000Z

## Abstract

Modern reasoning models continue generating long after the answer is already determined. Across five model configurations, two families, and three benchmarks, we find that 52--88% of chain-of-thought tokens are produced after the answer is recoverable from a partial prefix. This post-commitment generation reveals a structural phenomenon: the detection-extraction gap. Free continuations from early prefixes recover the correct answer even at 10% of the trace, while forced extraction fails on 42% of these cases. The answer is recoverable from the model state, yet prompt-conditioned decoding fails to extract it. We formalize this mismatch via a total-variation bound between free and forced continuation distributions, yielding quantitative estimates of suffix-induced shift. Exploiting this asymmetry, we propose Black-box Adaptive Early Exit (BAEE), which uses free continuations for both detection and extraction, truncating 70--78% of serial generation while improving accuracy by 1--5pp across all models. For thinking-mode models, early exit prevents post-commitment overwriting, yielding gains of up to 5.8pp; a cost-optimized variant achieves 68--73% reduction at a median of 9 API calls. Code is available at https://github.com/EdWangLoDaSc/know2say.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Question Answering on GPQA-D (∆Acc, CR)](<https://www.sota2.com/research/sota/question-answering-on-gpqa-d-acc-cr.md>): Question Answering; GPQA D; Coverage Rate: 51.

- [Mathematical Reasoning on MATH 500 (∆Acc, CR)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-math-500-acc-cr.md>): Mathematical Reasoning; MATH 500; Completion Rate: 32.

- [Code Generation on HumanEval (test)](<https://www.sota2.com/research/sota/code-generation-on-humaneval-test.md>): Code Generation; HumanEval (test); Pass@1: not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.06613>)

[PDF](<https://arxiv.org/pdf/2604.06613.pdf>)