Saved in:
Bibliographic Details
Main Authors: Singh, Harman, Li, Xiuyu, Sareen, Kusha, Maheswaran, Monishwaran, Tan, Sijun, Wu, Xiaoxia, Wang, Junxiong, Ariyak, Alpay, Wu, Qingyang, Khaki, Samir, Tiwari, Rishabh, Lian, Long, Lu, Yucheng, Li, Boyi, Suhr, Alane, Athiwaratkun, Ben, Keutzer, Kurt
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.04304
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914368695828480
author Singh, Harman
Li, Xiuyu
Sareen, Kusha
Maheswaran, Monishwaran
Tan, Sijun
Wu, Xiaoxia
Wang, Junxiong
Ariyak, Alpay
Wu, Qingyang
Khaki, Samir
Tiwari, Rishabh
Lian, Long
Lu, Yucheng
Li, Boyi
Suhr, Alane
Athiwaratkun, Ben
Keutzer, Kurt
author_facet Singh, Harman
Li, Xiuyu
Sareen, Kusha
Maheswaran, Monishwaran
Tan, Sijun
Wu, Xiaoxia
Wang, Junxiong
Ariyak, Alpay
Wu, Qingyang
Khaki, Samir
Tiwari, Rishabh
Lian, Long
Lu, Yucheng
Li, Boyi
Suhr, Alane
Athiwaratkun, Ben
Keutzer, Kurt
contents Test-time scaling for complex reasoning tasks shows that leveraging inference-time compute, by methods such as independently sampling and aggregating multiple solutions, results in significantly better task outcomes. However, a critical bottleneck is verification: sampling is only effective if correct solutions can be reliably identified among candidates. While existing approaches typically evaluate candidates independently via scalar scoring, we demonstrate that models are substantially stronger at pairwise self-verification. Leveraging this insight, we introduce $V_1$, a framework that unifies generation and verification through efficient pairwise ranking. $V_1$ comprises two components: $V_1$-Infer, an uncertainty-guided algorithm using a tournament-based ranking that dynamically allocates self-verification compute to candidate pairs whose relative correctness is most uncertain; and $V_1$-PairRL, an RL framework that jointly trains a single model as both generator and pairwise self-verifier, ensuring the verifier adapts to the generator's evolving distribution. On code generation (LiveCodeBench, CodeContests, SWE-Bench) and math reasoning (AIME, HMMT) benchmarks, $V_1$-Infer improves Pass@1 by up to $10%$ over pointwise verification and outperforms recent test-time scaling methods while being significantly more efficient. Furthermore, $V_1$-PairRL achieves $7$--$9%$ test-time scaling gains over standard RL and pointwise joint training, and improves base Pass@1 by up to 8.7% over standard RL in a code-generation setting.
format Preprint
id arxiv_https___arxiv_org_abs_2603_04304
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle $V_1$: Unifying Generation and Self-Verification for Parallel Reasoners
Singh, Harman
Li, Xiuyu
Sareen, Kusha
Maheswaran, Monishwaran
Tan, Sijun
Wu, Xiaoxia
Wang, Junxiong
Ariyak, Alpay
Wu, Qingyang
Khaki, Samir
Tiwari, Rishabh
Lian, Long
Lu, Yucheng
Li, Boyi
Suhr, Alane
Athiwaratkun, Ben
Keutzer, Kurt
Computation and Language
Test-time scaling for complex reasoning tasks shows that leveraging inference-time compute, by methods such as independently sampling and aggregating multiple solutions, results in significantly better task outcomes. However, a critical bottleneck is verification: sampling is only effective if correct solutions can be reliably identified among candidates. While existing approaches typically evaluate candidates independently via scalar scoring, we demonstrate that models are substantially stronger at pairwise self-verification. Leveraging this insight, we introduce $V_1$, a framework that unifies generation and verification through efficient pairwise ranking. $V_1$ comprises two components: $V_1$-Infer, an uncertainty-guided algorithm using a tournament-based ranking that dynamically allocates self-verification compute to candidate pairs whose relative correctness is most uncertain; and $V_1$-PairRL, an RL framework that jointly trains a single model as both generator and pairwise self-verifier, ensuring the verifier adapts to the generator's evolving distribution. On code generation (LiveCodeBench, CodeContests, SWE-Bench) and math reasoning (AIME, HMMT) benchmarks, $V_1$-Infer improves Pass@1 by up to $10%$ over pointwise verification and outperforms recent test-time scaling methods while being significantly more efficient. Furthermore, $V_1$-PairRL achieves $7$--$9%$ test-time scaling gains over standard RL and pointwise joint training, and improves base Pass@1 by up to 8.7% over standard RL in a code-generation setting.
title $V_1$: Unifying Generation and Self-Verification for Parallel Reasoners
topic Computation and Language
url https://arxiv.org/abs/2603.04304