Research

Paper

TESTING March 04, 2026

$V_1$: Unifying Generation and Self-Verification for Parallel Reasoners

Authors

Harman Singh, Xiuyu Li, Kusha Sareen, Monishwaran Maheswaran, Sijun Tan, Xiaoxia Wu, Junxiong Wang, Alpay Ariyak, Qingyang Wu, Samir Khaki, Rishabh Tiwari, Long Lian, Yucheng Lu, Boyi Li, Alane Suhr, Ben Athiwaratkun, Kurt Keutzer

Abstract

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.

Metadata

arXiv ID: 2603.04304
Provider: ARXIV
Primary Category: cs.CL
Published: 2026-03-04
Fetched: 2026-03-05 06:06

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.04304v1</id>\n    <title>$V_1$: Unifying Generation and Self-Verification for Parallel Reasoners</title>\n    <updated>2026-03-04T17:22:16Z</updated>\n    <link href='https://arxiv.org/abs/2603.04304v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.04304v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>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.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-03-04T17:22:16Z</published>\n    <arxiv:primary_category term='cs.CL'/>\n    <author>\n      <name>Harman Singh</name>\n    </author>\n    <author>\n      <name>Xiuyu Li</name>\n    </author>\n    <author>\n      <name>Kusha Sareen</name>\n    </author>\n    <author>\n      <name>Monishwaran Maheswaran</name>\n    </author>\n    <author>\n      <name>Sijun Tan</name>\n    </author>\n    <author>\n      <name>Xiaoxia Wu</name>\n    </author>\n    <author>\n      <name>Junxiong Wang</name>\n    </author>\n    <author>\n      <name>Alpay Ariyak</name>\n    </author>\n    <author>\n      <name>Qingyang Wu</name>\n    </author>\n    <author>\n      <name>Samir Khaki</name>\n    </author>\n    <author>\n      <name>Rishabh Tiwari</name>\n    </author>\n    <author>\n      <name>Long Lian</name>\n    </author>\n    <author>\n      <name>Yucheng Lu</name>\n    </author>\n    <author>\n      <name>Boyi Li</name>\n    </author>\n    <author>\n      <name>Alane Suhr</name>\n    </author>\n    <author>\n      <name>Ben Athiwaratkun</name>\n    </author>\n    <author>\n      <name>Kurt Keutzer</name>\n    </author>\n  </entry>"
}