FineVerify: Scaling Test-Time Compute with Fine-Grained Self-Verification for Agentic Search

Hugging Face Daily Papers Papers
agentic-search self-verification test-time-compute fine-grained language-model-agents benchmarks

Summary

FineVerify is a self-verification framework for agentic search that decomposes questions into sub-questions, verifies sampled candidates, and selects the best one, achieving substantial accuracy improvements over baselines on multiple benchmarks, including enabling GPT-5-mini to surpass GPT-5 on BrowseComp-Plus.

Agentic search requires language model agents to explore many sources and answer complex information-seeking questions. Scaling test-time compute is a promising way to improve these agents, but current approaches can fail, because correct answers are often sparse and score-based selection depends on model calibration. We propose FineVerify, a fine-grained self-verification framework that decomposes each question into checkable sub-questions, verifies sampled candidates against each sub-question, and selects the candidate with the highest aggregated score. This per-check structure turns selection into simpler local judgments and produces scores under the same explicit criteria. Across four agentic search benchmarks and two models, FineVerify consistently outperforms standard scaling baselines. With only four sampled trajectories, it improves GPT-5-mini by 8.2 accuracy points and Gemini-3-flash by 5.6% on average. With 12 samples, FineVerify enables GPT-5-mini to surpass frontier GPT-5 on BrowseComp-Plus. Beyond accuracy, FineVerify produces interpretable verification traces that help audit benchmark errors, suggesting broader applications for inspecting agentic search systems. Code and data are available at https://github.com/XuZhao0/fineverify
Original Article
View Cached Full Text

Cached at: 06/02/26, 03:36 PM

Paper page - FineVerify: Scaling Test-Time Compute with Fine-Grained Self-Verification for Agentic Search

Source: https://huggingface.co/papers/2606.00660

Abstract

FineVerify is a self-verification framework for agentic search that improves accuracy through decomposed sub-question checking and trajectory selection.

Agentic searchrequireslanguage model agentsto explore many sources and answer complex information-seeking questions. Scalingtest-time computeis a promising way to improve these agents, but current approaches can fail, because correct answers are often sparse andscore-based selectiondepends on model calibration. We propose FineVerify, afine-grained self-verificationframework that decomposes each question intocheckable sub-questions, verifiessampled candidatesagainst each sub-question, and selects the candidate with the highestaggregated score. This per-check structure turns selection into simpler local judgments and produces scores under the same explicit criteria. Across fouragentic searchbenchmarks and two models, FineVerify consistently outperforms standard scaling baselines. With only four sampled trajectories, it improves GPT-5-mini by 8.2 accuracy points and Gemini-3-flash by 5.6% on average. With 12 samples, FineVerify enables GPT-5-mini to surpass frontier GPT-5 onBrowseComp-Plus. Beyond accuracy, FineVerify produces interpretable verification traces that help audit benchmark errors, suggesting broader applications for inspectingagentic searchsystems. Code and data are available at https://github.com/XuZhao0/fineverify

View arXiv pageView PDFGitHub2Add to collection

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.00660 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.00660 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.00660 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

AgentV-RL: Scaling Reward Modeling with Agentic Verifier

arXiv cs.CL

AgentV-RL introduces an Agentic Verifier framework that enhances reward modeling through bidirectional verification with forward and backward agents augmented with tools, achieving 25.2% improvement over state-of-the-art ORMs. The approach addresses error propagation and grounding issues in verifiers for complex reasoning tasks through multi-turn deliberative processes combined with reinforcement learning.

VeriGate: Verifier-Gated Step-Level Supervision for GRPO

arXiv cs.LG

VeriGate extends GRPO with verifier-gated step-level supervision, providing fine-grained credit assignment when verifier rewards are degenerate. It achieves substantial accuracy improvements on reasoning benchmarks for 1.5B and 7B models.

Agentic Proving for Program Verification

arXiv cs.AI

This paper evaluates Claude Code in an agentic proving framework on the Clever benchmark for program verification, achieving over 98% success in specification generation and end-to-end verification, revealing that existing benchmarks may be insufficient for evaluating modern agentic provers.

Verifying Agentic Development at Scale (8 minute read)

TLDR AI

Cognition's Ido Pesok shares lessons from building autonomous end-to-end testing into Devin, noting that for the first time, more Devin sessions are now triggered asynchronously than interactively, making verified-before-merge results a hard requirement rather than a nicety.