Pith. sign in

REVIEW 3 major objections 5 minor 53 references

SearchMaster: Grounded and Regulated Self-Play for Search Agents

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read SearchMaster claims that a 9-billion-parameter model can bootstrap deep-search ability by generating, solving, and verifying its own tasks in a local search environment.

desk verdict A genuinely novel self-play recipe for search agents with credible ablations; the self-referential Verifier is the main uncertainty, but the results deserve serious referee time. read the letter →

arxiv 2608.01822 v1 pith:C4II5D5D submitted 2026-08-03 cs.AI

Wentao Tan , Qiong Cao , Jiaqi Wang , Nan Duan This is my paper · ORCID
classification cs.AI
keywords searchagentsself-playevidencechainsearch-depthrewardover-openingpenaltyGRPOdeepmulti-hopquestionanswering
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

SearchMaster claims that a single language model can train itself into a capable deep-search agent using only tasks it generates, solves, and verifies in a local search environment, with no human-labeled question-answer pairs or expert demonstrations. The paper identifies three ways self-play can go wrong: questions that look multi-hop but are answerable from one document, difficulty estimates based on success rate rather than search depth, and trajectories that open many documents without gathering targeted evidence. It then adds a control for each failure mode. On six deep-search benchmarks, the framework raises a 9-billion-parameter backbone from 38.19% to 51.52% average accuracy, including a 30.1-point jump on BrowseComp-Plus. If the result holds, search-agent skill can be bootstrapped from a model's own exploration, removing one of the main data bottlenecks in deep-research systems.

What carries the argument

The machinery is a closed self-play loop in which one shared policy plays two roles. As Proposer, it starts from a seed document, builds an explicit evidence chain using search, open, and find tools, and writes a question plus a reference answer derived from the whole chain. As Solver, it receives only the question and produces tool-use rollouts; a frozen Verifier, an untrained copy of the initial model, scores each answer correct or incorrect. Four filters remove shallow, malformed, invalid, or parametric-knowledge-answerable candidates before Solver rollouts. Rewards combine correctness with the Search-Depth Reward, which uses the minimum search depth among correct rollouts to measure task difficulty, and the Over-Opening Penalty, which clips the open-to-search ratio into a linear penalty. GRPO turns these rewards into token-level advantages, and the same policy is updated on both Proposer and Solver samples.

What would settle it

Take a sample of SearchMaster-generated tasks with their Verifier scores, have independent annotators or a stronger external judge relabel correctness, and check whether the accepted answers are actually correct; if the Verifier accepts a substantial share of wrong answers, the benchmark gains would not reflect genuine search ability.

Watch

Extended reading notes

Core claim

The paper reports that the failure modes of search self-play can be corrected by making three quantities explicit. The Evidence-Chain Generator forces task proposals to build a visible chain of evidence across documents before asking a question, which raises the measured fraction of genuinely multi-hop tasks from 24.2% to 78.6% while cutting invalid tasks from 28.7% to 6.4%. The Search-Depth Reward scores a task by the minimum number of unique searches among successful solutions, keeping retained tasks search-intensive as the solver improves. The Over-Opening Penalty subtracts a penalty proportional to the open-to-search ratio, preventing the policy from drifting into repeated document openings. With these controls, verified Proposer and Solver rollouts are optimized jointly with GRPO, and the resulting policy improves the untrained backbone by 30.1 points on BrowseComp-Plus and improves all five live-web benchmarks, despite never training on the live web.

Load-bearing premise

The load-bearing premise is that the frozen Verifier, a copy of the initial model, can reliably tell correct answers from plausible wrong ones on self-generated tasks; if it accepts wrong reference answers, training rewards could push the policy toward self-consistent errors rather than genuine search ability.

Editorial extensions

If this is right

  • Search-agent training can proceed with no human-written questions or expert trajectories, using only an indexed document collection and a base model.
  • Search behavior learned offline transfers to live, open-web benchmarks, so local self-play may be a viable substitute for expensive web-scale reinforcement learning.
  • Task difficulty measured by minimum successful search depth prevents the training distribution from collapsing into shallow lookup tasks as the solver improves.
  • A 9-billion-parameter model trained this way can exceed much larger open and proprietary systems on at least one deep-search benchmark, suggesting model scale is not the only driver of search skill.
  • The open-to-search penalty offers a reusable control for tool-use efficiency in other agentic reinforcement-learning settings.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Going beyond the paper, the Over-Opening Penalty is stated generically enough that it could be transferred to other tool-use agents, such as coding agents or database-querying agents, where repeated opening of the same resource is a known failure mode.
  • A natural stress test would be to remove the parametric-knowledge filter and see whether the frozen Verifier's judgments degrade when reference answers are less grounded in retrieved evidence.
  • The evidence-chain quality could be cheaply monitored during training with an independent lightweight classifier, potentially replacing part of the Verifier's validity-filtering role and reducing compute.
  • Because the paper's difficulty signal uses only the minimum successful search depth, a task with one lucky shallow solution is treated as easy even if most successful rollouts are deep; an alternative that also accounts for the distribution of depths might change which tasks are retained.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes SearchMaster, a self-play framework for training an LLM-based search agent without human-labeled QA pairs or expert demonstrations. A single Qwen3.5-9B policy acts as both Proposer and Solver inside a local search environment. The Proposer builds explicit evidence chains (ECG) and generates tasks; four filters remove invalid candidates; the Solver produces multiple rollouts; a frozen copy of the initial model serves as Verifier, scoring correctness; a Search-Depth Reward (SDR) assigns task difficulty by the minimum successful search depth; and an Over-Opening Penalty (OOP) regularizes the open-to-search ratio. The policy is updated with GRPO using the resulting rewards. On six deep-search benchmarks, the method improves the backbone from 38.19% to 51.52% average accuracy, including a 30.1-point gain on BrowseComp-Plus, and ablations show monotonic gains from ECG, SDR, and OOP.

Significance. If the results hold, this is a notable advance: it shows that a 9B model can bootstrap competitive deep-search ability from purely self-generated tasks in a local corpus, surpassing much larger open and proprietary baselines on BrowseComp-Plus and transferring to live-web benchmarks. The paper also provides a useful decomposition of the self-play failure modes and an ablation that demonstrates each mechanism contributes. The explicit release commitment for code, checkpoints, and training data is a strength. However, the central causal claim that the training signal is 'grounded and regulated' depends on the reliability of the frozen Verifier, which is a copy of the initial model and is not audited in the paper. The final benchmark numbers are not circular because they use external judges and held-out benchmarks, but the training signal is self-referential. This concern, together with the absence of variance estimates, is the main load-bearing uncertainty.

major comments (3)
  1. [Solver Rollouts and Search Depth; Eq. (6)] The frozen Verifier V is the sole correctness signal in training, and V is a frozen copy of the initial Qwen3.5-9B model. The reference answer y*_j is generated by the Proposer, and the validity, parametric-knowledge, and correctness filters all use the same Verifier. Since the Solver base reward is bS_i = z_i and the Proposer reward depends on C = sum_i z_i, a systematically lenient Verifier could train the policy toward self-consistent but factually incorrect answers. The external benchmark evaluation means the headline numbers are not circular, but the causal claim that gains come from 'grounded and regulated' self-play requires direct evidence. Please add a Verifier audit: sample retained training tasks and have human annotators or a strong external judge label whether the reference answers are correct and whether the Verifier's z_i agrees with external correctness, reporting precision, recall, and agreement. A small training comparison using an external verifier would also directly test whether the gains are an artifact of the self-referential signal. This is the main load-bearing issue in the paper.
  2. [Table 1 and Table 2] All reported accuracies are single-run numbers with no error bars, confidence intervals, or significance tests. Several evaluation sets are small (GAIA has 103 instances, SEAL-0 has 111, XBench has 100), so differences of a few points may be within noise. The 30.1-point gain on BrowseComp-Plus is large and unlikely to be noise, but the average online gain of 13.3 points is supported by variable per-benchmark improvements. Please report results over at least three training seeds or provide bootstrapped confidence intervals, and do the same for the ablation in Table 2. This is necessary to assess the magnitude and stability of the claimed improvements.
  3. [Evidence-Chain Task Generation; Table 3] The ECG diagnostic in Table 3 labels task quality using GLM-5, another LLM judge, rather than human annotation. The paper reports a large increase in 'True Multi-Hop' (from 24.2% to 78.6%) and a drop in 'Invalid' tasks, but without human agreement or error analysis it is unclear how reliable these labels are. Since the central claim is that ECG reduces pseudo multi-hop tasks, please include a human-validated sample (e.g., 200 tasks per condition) or report agreement between GLM-5 and human judges. This would strengthen the behavioral evidence beyond the final accuracy numbers.
minor comments (5)
  1. [Figure 3] The caption says 'range' but does not define whether it is min-max, interquartile, or standard deviation. Please specify the exact definition in the caption or text.
  2. [Table 1] The Avg. column is computed only for the two rows with all six benchmark entries; external baselines with missing entries show dashes. Please state this explicitly in the table caption so the reader does not compare Avg. values across rows with different support.
  3. [Eq. (10)] The notation in Eq. (10) uses gamma_t for the per-token importance ratio and also for the clip argument; the definition appears only in the text after the equation. Please define gamma_t before the equation or restructure the expression for readability.
  4. [Method: Quality Filters] The parametric-knowledge filter is described only as 'removes questions the Verifier can answer without tools.' Please specify how this is operationalized: for example, whether the Verifier is prompted to answer q_j directly without tool observations and what threshold or decision rule is used.
  5. [Conclusion] The limitation about operating in a local search environment is acknowledged, but the paper does not discuss the risk that the OpenResearcher corpus may have distributional overlap with current web content or that the offline environment's fixed documents could become stale for live-benchmark transfer. A sentence on corpus diversity and timeliness would be useful.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the claimed benchmark gains are measured externally, so the internal self-play reward loop does not reduce the headline result to its inputs.

full rationale

SearchMaster's derivation chain is a closed RL loop: the Proposer generates candidate tasks, the Solver rolls out answers, a frozen Verifier scores correctness, and GRPO updates the shared policy. The Verifier is indeed self-referential in the sense that it is a frozen copy of the initial model judging reference answers generated by the same policy family, but the paper does not present the Verifier's judgments as the claimed outcome. The headline claims are accuracy numbers on six external benchmarks (BrowseComp-Plus, BrowseComp, GAIA, SEAL-0, WebWalkerQA, XBench), scored by GPT-5 against ground-truth answers, rather than by the Verifier or by the training rewards. No equation in the paper reduces a benchmark result to a training signal by construction. The SDR and OOP analyses track hmin and open/search ratio, which are the exact quantities those rewards are designed to push; these are reward-conditioning sanity checks rather than independent predictions, so observing that SDR retains high-hmin tasks or that OOP lowers the open/search ratio is not circular derivation of the main empirical claim. The central uncertainties, such as whether the frozen Verifier reliably accepts only correct reference answers or whether the reported single-run improvements have error bars, are empirical validity concerns rather than circularity. No load-bearing self-citation chain or imported uniqueness theorem appears; the method is evaluated against external benchmarks and prior systems. Therefore the appropriate finding is no significant circularity.

Assumptions & free parameters 10 free parameters · 5 assumptions · 0 invented entities

The central claim depends on a chain of unverified assumptions: the frozen initial Verifier is reliable, self-generated reference answers are factually correct enough, the OpenResearcher corpus and Qwen3 embedding retriever support transfer to live web, and GPT-5 scoring plus external baseline numbers are trustworthy. The free parameters are hand-chosen reward and sampling hyperparameters; none is fitted to benchmark data, but no sensitivity analysis is reported. No invented physical or conceptual entities are introduced; ECG, SDR, and OOP are training procedures, not entities.

free parameters (10)
  • r0 (base reward) = 0.2
    Reward for tasks that are too hard or too easy (C=0 or C=K_S); hand-set in the reward design section.
  • Hsdr (SDR saturation depth) = 10
    Target search depth at which the Search-Depth Reward saturates in Eq. (4).
  • alpha_oop = 1.5
    Open-to-search ratio where the over-opening penalty begins in Eq. (7).
  • beta_oop = 2.5
    Open-to-search ratio where the over-opening penalty reaches its maximum in Eq. (7).
  • lambda_oop = 0.5
    Coefficient scaling the over-opening penalty in Equations (8) and (9).
  • K_P, K_S = 8 each
    Number of Proposer and Solver rollouts per seed or task; affects reward estimation and batch size.
  • Tool-use filter threshold
    Threshold for 'too few search or open actions' is not specified in the main text and is deferred to the Supplementary Material.
  • Retained seeds per iteration = 64
    Seed-level quality gate keeps 64 qualified seeds per iteration for training.
  • Training iterations = 20
    The model trains on 20 x 64 = 1,280 seed documents; no convergence analysis is reported.
  • GRPO hyperparameters = LR 1e-6, clip range (0.2, 0.28), KL weight 0.001
    Standard GRPO settings chosen without sensitivity analysis.
assumptions (5)
  • domain assumption Frozen Verifier reliability
    The initial Qwen3.5-9B is trusted to judge task validity, filter parametric-knowledge questions, and score Solver answer correctness; no external validation of Verifier judgments is reported.
  • domain assumption Self-generated reference answers are correct
    The Proposer's y*_j is used as ground truth for Solver rewards; there is no independent factual check that y*_j is correct, only that the Proposer's trace supports it.
  • domain assumption Corpus and retriever transfer
    Training on the OpenResearcher offline corpus with Qwen3-Embedding-8B and FAISS transfers to BrowseComp-Plus and live web benchmarks; no corpus overlap or domain-gap analysis is shown.
  • domain assumption GPT-5 judge accuracy
    All final benchmark answers are scored by GPT-5 using a prompt; no human agreement study or judge calibration is provided.
  • domain assumption External baselines comparability
    Baseline numbers in Table 1 are copied from source papers with different evaluation harnesses, context lengths, and judges; comparability is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SearchMaster: Grounded and Regulated Self-Play for Search Agents." pith.science (2026). https://pith.science/paper/C4II5D5D

@misc{pith2026260801822,
  author       = {Pith},
  title        = {Pith review of: SearchMaster: Grounded and Regulated Self-Play for Search Agents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C4II5D5D}},
  note         = {Machine review of arXiv:2608.01822}
}
read the original abstract

Training LLM-based search agents requires high-quality search data: tasks that demand genuine multi-hop retrieval and trajectories that use search tools effectively. Existing pipelines often depend on human-written tasks, expert demonstrations, or stronger teacher models. We present SearchMaster, a self-play framework that trains a single LLM from search tasks it generates, solves, and verifies in a local search environment. The key challenge is that self-generated tasks and rollouts can yield misleading signals: pseudo multi-hop questions, success-rate difficulty estimates that ignore search depth, and rollouts with excessive opening but little targeted evidence acquisition. SearchMaster addresses these failure modes with three controls. An Evidence-Chain Generator (ECG) grounds task generation in explicit cross-document evidence chains to reduce pseudo multi-hop questions. A Search-Depth Reward (SDR) scores task difficulty by the search depth of successful rollouts rather than success rate alone, keeping retained tasks search-intensive. An Over-Opening Penalty (OOP) regulates tool use by discouraging excessive document opening, avoiding long but shallow browsing. Verified Proposer and Solver rollouts are then jointly optimized with GRPO. Across six deep-search benchmarks, SearchMaster improves a Qwen3.5-9B backbone from 38.19% to 51.52% average accuracy, with a 30.1-point gain on BrowseComp-Plus. These results show that grounded and regulated self-play can provide effective search-agent training data without human-labeled QA pairs or expert demonstrations. The code is available at https://github.com/WentaoTan/SearchMaster.

Figures

Figures reproduced from arXiv: 2608.01822 by the authors.

Figure 1
Figure 1. SearchMaster grounds tasks and regulates behav [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the SearchMaster self-play loop. A shared policy [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Average open/search ratio of the Proposer and the [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (1 more)
Figure 3
Figure 3. Figure 3: Search depth hmin of the training tasks over self￾play. Lines show the mean and bands show the range across tasks. tion generation. Together, these effects turn loosely related browsing into task-relevant cross-document dependencies. Does SDR Keep the Training Tasks Se…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 6 canonical work pages

  1. [1]

    Findings of the Association for Computational Linguistics: ACL 2026 , pages=

    AutoSearch: Adaptive Search Depth for Efficient Agentic RAG via Reinforcement Learning , author=. Findings of the Association for Computational Linguistics: ACL 2026 , pages=

    work page 2026

  2. [2]

    arXiv preprint arXiv:2112.09332 , year=

    Webgpt: Browser-assisted question-answering with human feedback , author=. arXiv preprint arXiv:2112.09332 , year=

    arXiv

  3. [3]

    arXiv preprint arXiv:2210.03629 , year=

    React: Synergizing reasoning and acting in language models , author=. arXiv preprint arXiv:2210.03629 , year=

    arXiv

  4. [4]

    arXiv preprint arXiv:2603.20278 , year=

    Openresearcher: A fully open pipeline for long-horizon deep research trajectory synthesis , author=. arXiv preprint arXiv:2603.20278 , year=

    arXiv

  5. [5]

    arXiv preprint arXiv:2507.02592 , year=

    Websailor: Navigating super-human reasoning for web agent , author=. arXiv preprint arXiv:2507.02592 , year=

    arXiv

  6. [6]

    arXiv preprint arXiv:2503.09516 , year=

    Search-r1: Training llms to reason and leverage search engines with reinforcement learning , author=. arXiv preprint arXiv:2503.09516 , year=

    arXiv

  7. [7]

    arXiv preprint arXiv:2503.05592 , year=

    R1-searcher: Incentivizing the search capability in llms via reinforcement learning , author=. arXiv preprint arXiv:2503.05592 , year=

    arXiv

  8. [8]

    Findings of the Association for Computational Linguistics: ACL 2026 , pages=

    R-search: Empowering llm reasoning with search via multi-reward reinforcement learning , author=. Findings of the Association for Computational Linguistics: ACL 2026 , pages=

    work page 2026

Show all 53 references
  1. [9]

    arXiv preprint arXiv:2505.15107 , year=

    Stepsearch: Igniting llms search ability via step-wise proximal policy optimization , author=. arXiv preprint arXiv:2505.15107 , year=

    arXiv

  2. [10]

    Advances in Neural Information Processing Systems , volume=

    Search and refine during think: Facilitating knowledge refinement for improved retrieval-augmented reasoning , author=. Advances in Neural Information Processing Systems , volume=

  3. [11]

    Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    Deepresearcher: Scaling deep research via reinforcement learning in real-world environments , author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    2025

  4. [12]

    arXiv preprint arXiv:2602.14234 , year=

    Redsearcher: A scalable and cost-efficient framework for long-horizon search agents , author=. arXiv preprint arXiv:2602.14234 , year=

  5. [13]

    arXiv preprint arXiv:2601.03743 , year=

    O-Researcher: An Open Ended Deep Research Model via Multi-Agent Distillation and Agentic RL , author=. arXiv preprint arXiv:2601.03743 , year=

  6. [14]

    arXiv preprint arXiv:2605.22511 , year=

    Search-E1: Self-Distillation Drives Self-Evolution in Search-Augmented Reasoning , author=. arXiv preprint arXiv:2605.22511 , year=

    arXiv

  7. [15]

    Advances in Neural Information Processing Systems , volume=

    Webthinker: Empowering large reasoning models with deep research capability , author=. Advances in Neural Information Processing Systems , volume=

  8. [16]

    Advances in Neural Information Processing Systems , volume=

    Webdancer: Towards autonomous information seeking agency , author=. Advances in Neural Information Processing Systems , volume=

  9. [17]

    Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    Webagent-r1: Training web agents via end-to-end multi-turn reinforcement learning , author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    2025

  10. [18]

    arXiv preprint arXiv:2509.06501 , year=

    Webexplorer: Explore and evolve for training long-horizon web agents , author=. arXiv preprint arXiv:2509.06501 , year=

  11. [19]

    arXiv preprint arXiv:2508.05004 , year=

    R-zero: Self-evolving reasoning llm from zero data , author=. arXiv preprint arXiv:2508.05004 , year=

    arXiv

  12. [20]

    Advances in Neural Information Processing Systems , volume=

    Absolute zero: Reinforced self-play reasoning with zero data , author=. Advances in Neural Information Processing Systems , volume=

  13. [21]

    arXiv preprint arXiv:2602.21320 , year=

    Tool-r0: Self-evolving llm agents for tool-learning from zero data , author=. arXiv preprint arXiv:2602.21320 , year=

  14. [22]

    arXiv preprint arXiv:2510.18821 , year=

    Search self-play: Pushing the frontier of agent capability without supervision , author=. arXiv preprint arXiv:2510.18821 , year=

    arXiv

  15. [23]

    Zero: Self-Evolving Search Agents without Training Data , author=

    Dr. Zero: Self-Evolving Search Agents without Training Data , author=. arXiv preprint arXiv:2601.07055 , year=

    arXiv

  16. [24]

    arXiv preprint arXiv:2402.03300 , year=

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models , author=. arXiv preprint arXiv:2402.03300 , year=

    arXiv

  17. [25]

    arXiv preprint arXiv:2505.09388 , year=

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    arXiv

  18. [26]

    arXiv preprint arXiv:2601.03267 , year=

    Openai gpt-5 system card , author=. arXiv preprint arXiv:2601.03267 , year=

    arXiv

  19. [27]

    2025 , institution =

    2025

  20. [28]

    2025 , howpublished =

    2025

  21. [29]

    arXiv preprint arXiv:2602.15763 , year=

    Glm-5: from vibe coding to agentic engineering , author=. arXiv preprint arXiv:2602.15763 , year=

    arXiv

  22. [30]

    arXiv preprint arXiv:2412.19437 , year=

    Deepseek-v3 technical report , author=. arXiv preprint arXiv:2412.19437 , year=

    arXiv

  23. [31]

    arXiv preprint arXiv:2509.10446 , year=

    Deepdive: Advancing deep search agents with knowledge graphs and multi-turn rl , author=. arXiv preprint arXiv:2509.10446 , year=

  24. [32]

    arXiv preprint arXiv:2511.11793 , year=

    Mirothinker: Pushing the performance boundaries of open-source research agents via model, context, and interactive scaling , author=. arXiv preprint arXiv:2511.11793 , year=

    arXiv

  25. [33]

    arXiv preprint arXiv:2510.24701 , year=

    Tongyi deepresearch technical report , author=. arXiv preprint arXiv:2510.24701 , year=

    arXiv

  26. [34]

    arXiv preprint arXiv:2508.09129 , year=

    Browsemaster: Towards scalable web browsing via tool-augmented programmatic agent pair , author=. arXiv preprint arXiv:2508.09129 , year=

    arXiv

  27. [35]

    arXiv preprint arXiv:2510.24697 , year=

    Webleaper: Empowering efficiency and efficacy in webagent via enabling info-rich seeking , author=. arXiv preprint arXiv:2510.24697 , year=

  28. [36]

    arXiv preprint arXiv:2509.13305 , year=

    Websailor-v2: Bridging the chasm to proprietary agents via synthetic data and scalable reinforcement learning , author=. arXiv preprint arXiv:2509.13305 , year=

  29. [37]

    arXiv preprint arXiv:2601.18467 , year=

    OffSeeker: Online Reinforcement Learning Is Not All You Need for Deep Research Agents , author=. arXiv preprint arXiv:2601.18467 , year=

  30. [38]

    arXiv preprint arXiv:2507.15061 , year=

    Webshaper: Agentically data synthesizing via information-seeking formalization , author=. arXiv preprint arXiv:2507.15061 , year=

    arXiv

  31. [39]

    arXiv preprint arXiv:2508.10925 , year=

    gpt-oss-120b & gpt-oss-20b model card , author=. arXiv preprint arXiv:2508.10925 , year=

    arXiv

  32. [40]

    arXiv preprint arXiv:2508.06471 , year=

    Glm-4.5: Agentic, reasoning, and coding (arc) foundation models , author=. arXiv preprint arXiv:2508.06471 , year=

    arXiv

  33. [41]

    arXiv preprint arXiv:2507.20534 , year=

    Kimi k2: Open agentic intelligence , author=. arXiv preprint arXiv:2507.20534 , year=

    arXiv

  34. [42]

    arXiv preprint arXiv:2504.12516 , year=

    Browsecomp: A simple yet challenging benchmark for browsing agents , author=. arXiv preprint arXiv:2504.12516 , year=

    arXiv

  35. [43]

    arXiv preprint arXiv:2506.13651 , year=

    xbench: Tracking agents productivity scaling with profession-aligned real-world evaluations , author=. arXiv preprint arXiv:2506.13651 , year=

    arXiv

  36. [44]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Webwalker: Benchmarking llms in web traversal , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  37. [45]

    arXiv preprint arXiv:2506.01062 , year=

    SealQA: Raising the Bar for Reasoning in Search-Augmented Language Models , author=. arXiv preprint arXiv:2506.01062 , year=

    arXiv

  38. [46]

    International Conference on Learning Representations , volume=

    Gaia: a benchmark for general ai assistants , author=. International Conference on Learning Representations , volume=

  39. [47]

    Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    BrowseComp-Plus: A Fair and Disentangled Evaluation Benchmark for Deep Search Agents , author=. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  40. [48]

    International conference on machine learning , pages=

    Retrieval augmented language model pre-training , author=. International conference on machine learning , pages=. 2020 , organization=

    2020

  41. [49]

    Advances in neural information processing systems , volume=

    Toolformer: Language models can teach themselves to use tools , author=. Advances in neural information processing systems , volume=

  42. [50]

    Advances in neural information processing systems , volume=

    Retrieval-augmented generation for knowledge-intensive nlp tasks , author=. Advances in neural information processing systems , volume=

  43. [51]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Rag-r1: Incentivizing the search and reasoning capabilities of llms through multi-query parallelism , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  44. [52]

    arXiv preprint arXiv:2505.16582 , year=

    O ^2 -Searcher: A Searching-based Agent Model for Open-Domain Open-Ended Question Answering , author=. arXiv preprint arXiv:2505.16582 , year=

    arXiv

  45. [53]

    2025 , howpublished =

    Zilin Zhu and Chengxing Xie and Xin Lv and slime Contributors , title =. 2025 , howpublished =

    2025

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.