Multiagent Systems · arXiv:2611.17041 v1 · submitted 12 Nov 2026 · 37 references
From Delegation to Coordination: Learning Efficient Multi-Agent Deep Research
The paper says
“Multi-agent deep-research systems use delegation to parallelize search and keep long tool traces out of the main agent's context.”
more of the abstract
Yet delegation helps only when task allocation and subtask execution are coordinated: the main agent must issue useful subtasks, while subagents must return reliable evidence. Existing training either optimizes only the main agent or broadcasts the same team advantage to all agents, which can reward misleading subagent reports and penalize useful ones. We introduce LEAD-RL (Learning Effective Agent Delegation with Reinforcement Learning), a joint main-agent-subagent training framework with contribution-aware credit assignment. A training-time rubric evaluates the research utility of each subagent report and gates its inherited team advantage, while final-answer correctness remains the only team reward; no token, step, or speedup reward is used. Across five benchmarks and two model families, LEAD-RL improves answer quality. On BrowseComp, it improves accuracy by 2.49 percentage points and reduces critical-path cost by 24.5% over joint training with shared team credit. During validation, accuracy increases by 6.78 points at approximately constant critical-path cost. These results show that contribution-aware credit assignment improves the quality-efficiency trade-off of multi-agent deep research.
from the abstract
Pith says
Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
Holds up, with conditions. LEAD-RL introduces a clean idea—rubric-gated credit assignment for subagent reports—and backs it with thorough experiments. The main soft spot is that the rubric judge's reliability is not independently validated, but the paper acknowledges this and the central claim holds up. This line is wrong
3 major objections · 7 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 37 references say.
The paper says, at §3.3, Eq. (5)-(6), Appendix A
The review did not quote this passage.
Pith says: Objection 1 of 3 · major
The contribution judge (DeepSeek V4 Pro) is used without independent validation of its accuracy, inter-rater reliability, or sensitivity to the specific prompts and evidence truncation. The paper acknowledges that confidence below 0.8 is mapped to neutral (Appendix A), but this does not establish that the judge's labels are correct or consistent.
more of objection 1
The central comparison in Table 3 conflates the credit-assignment mechanism with the specific labeling behavior of this judge. A proper ablation would compare LEAD-RL against a version that uses the same judge labels but applies them randomly or with a fixed gate (e.g., always neutral). Without such an ablation, the observed gains could be due to the judge's specific (possibly biased) labeling rather than the credit-assignment rule itself. The authors should either provide evidence of judge accuracy (e.g., human evaluation on a sample of reports) or include an ablation that controls for the judge's presence.
The paper says, at §4.4, Table 3
The review did not quote this passage.
Pith says: Objection 2 of 3 · major
The controlled comparison shows that LEAD-RL outperforms joint RL with shared team credit, but both RL variants underperform the SFT initialization in terms of accuracy (70.39% and 71.00% vs. 72.50%). This is a concerning pattern: the RL training appears to hurt performance unless the contribution-aware gating is used.
more of objection 2
The authors should discuss why main-only and shared-credit RL degrade accuracy relative to SFT, and whether this indicates a fundamental issue with the RL formulation (e.g., reward sparsity, instability) that LEAD-RL only partially mitigates. The fact that LEAD-RL recovers and slightly exceeds SFT accuracy is positive, but the degradation of the baselines raises questions about the robustness of the training pipeline.
The paper says, at §4.5, Figure 2, Appendix C
The review did not quote this passage.
A classifier read this sentence of the paper as acknowledging the concern (probability 0.62):
“Because both questions and trajectories vary across updates, these measurements expose the behavior optimized during training but are not estimates of generalization performance.”
Pith says: Objection 3 of 3 · major
The validation dynamics (Figure 2) show that LEAD-RL improves accuracy at approximately constant critical-path cost, while baselines increase cost substantially. However, Appendix C (Figure 3) reveals that LEAD-RL's critical-path cost begins to rise more rapidly after approximately 50 updates without a sustained increase in batch correctness.
more of objection 3
The authors note this late-stage behavior and suggest that checkpoint selection should consider the joint accuracy-efficiency frontier. This is an important caveat: the reported gains (2.49 points, 24.5% cost reduction) are from a specific checkpoint, and the method may not be stable over longer training. The authors should clarify whether the reported results in Table 3 correspond to the best checkpoint on the validation set or the final checkpoint, and whether the gains are robust to checkpoint selection.
7 minor objections
The paper says, at §1, Abstract
Pith says: minor objection 1 of 7
The abstract states 'no token, step, or speedup reward is used' but the method does use a KL penalty coefficient (βL_KL in Eq. 8), which is a form of regularization that could affect token-level behavior. This is not a major issue but the phrasing could be clarified.
The paper says, at §3.3, Eq. (7)
Pith says: minor objection 2 of 7
The normalization factor max(1, |I_j|) is used to prevent rollouts with more subagents from receiving disproportionately large aggregate updates. However, the denominator is the number of subagents with nonzero gated advantages, not the total number of subagents. This could incentivize the policy to produce more subagents with zero gated advantage (i.e., neutral or filtered out) to increase the per-subagent update. The authors should discuss this potential incentive.
show 5 more
The paper says, at §4.1, Table 1
Pith says: minor objection 3 of 7
The comparison with closed-source and large open-source models is informative but the paper correctly notes that external scores may use different scaffolds. The statement 'LEAD-RL (124B-A5.1B) outperforms DeepSeek V4 Flash (284B-A13B)' should be qualified: this is within the same swarm harness, but DeepSeek V4 Flash may perform differently under its own optimal harness. The paper already acknowledges this in §4.1 but the phrasing in §4.3 could be more precise.
The paper says, at Appendix A
Pith says: minor objection 4 of 7
The contribution rubric judge prompt is provided in full, which is good for reproducibility. However, the prompt includes the instruction 'Return JSON only' and specifies a schema. It would be helpful to report the average confidence scores and the proportion of labels mapped to neutral due to confidence < 0.8, to give readers a sense of how often the judge is uncertain.
The paper says, at §3.2
Pith says: minor objection 5 of 7
The local and global verifiers are described as 'role-conditioned invocations of the shared policy' and 'use no reference answer at inference'. It would be helpful to clarify whether these verifiers are also trained during RL or are fixed. If they are trained, how does their training interact with the main agent and subagent updates?
The paper says, at §4.2, Table 1
Pith says: minor objection 6 of 7
The table reports results for LEAD-RL on two model families, but the baseline Ling-3-Flash and Qwen3.6-35B-A3B results are from the same swarm harness. It would be helpful to also report the performance of these base models without any training (i.e., the SFT checkpoint) in the same table, to make the comparison more direct. Currently, the SFT initialization is only reported in Table 3 for BrowseComp.
The paper says, at Appendix D
Pith says: minor objection 7 of 7
The case study is illustrative but the paper correctly notes that it is 'not used in the aggregate evaluation' and the values are 'report contents rather than independently annotated ground truth'. This is appropriate, but the case study would be strengthened by including a failure case or an example where the rubric judge's classification is shown to be correct or incorrect.
What would settle it
Re-run the LEAD-RL training on BrowseComp with the contribution judge labels replaced by random labels (drawn from the same marginal distribution as the judge's labels, e.g., ~31% helpful, ~10% harmful, ~59% neutral) but still using the gating function of Eq. 6. If accuracy and critical-path cost remain similar to the reported LEAD-RL results, then the rubric's specific labels are not driving the improvement. If performance degrades to the level of joint RL with shared credit (or worse), then the judge's accuracy is essential. This test directly isolates the contribution of the judge's accuracy from the credit-assignment structure.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 13 Nov 2026
- Reader holds up, with conditions
- Skeptic agrees
- Editor wrote the verdict above
- Referee major revision: 3 major objections, 7 minor
Not run on this paper: the simulated authors' defence.
Pith did not check this review's objections against the paper.
The review in full
The review in brief, machine-written
The paper's core idea is simple and well-motivated: when training a multi-agent research system, don't give every subagent the same team reward. Instead, use a rubric to classify each subagent report as helpful, neutral, or harmful, and gate the team advantage accordingly. This is a genuine contribution—prior work either trained only the main agent or broadcast the same signal to all agents. The experiments are thorough: five benchmarks, two model families, controlled comparisons that isolate the RL stage, and validation dynamics that show the accuracy-efficiency trade-off improving during training without an explicit cost penalty. The paper also ships a clear formulation of critical-path cost and idealized speedup, which are useful evaluation tools.
The main soft spot is the reliance on a proprietary rubric judge (DeepSeek V4 Pro) without independent validation of its accuracy or inter-rater reliability. The paper acknowledges this indirectly—it maps low-confidence judgments to neutral, and it states that the mechanism remains a hypothesis rather than a directly measured effect—but the central comparison (Table 3) could conflate rubric quality with the credit-assignment mechanism. A proper ablation comparing LEAD-RL against a version that uses the same judge labels but applies them randomly or with a fixed gate would strengthen the claim. That said, the paper's own limitation statements are honest, and the concern is not fatal: the judge is used only for credit routing, not as a reward signal, and the results are consistent across two model families and five benchmarks.
Minor issues: no error bars or statistical significance tests for the main results, and code/data are not released. These are common in this area and don't undermine the core finding.
Who this is for: researchers working on multi-agent RL for language agents, especially those interested in credit assignment for delegation. The paper deserves a serious referee—the idea is clean, the experiments are extensive, and the limitations are acknowledged. I would accept peer review and recommend the editor send it out.
The strongest objection
The paper's central claim is that contribution-aware credit assignment (LEAD-RL) improves accuracy and reduces critical-path cost relative to joint RL with shared team credit. The key mechanism is the rubric-based gate (Eq. 6) that filters the team advantage for each subagent report. However, the rubric judge (DeepSeek V4 Pro) is used without any independent validation of its accuracy, inter-rater reliability, or sensitivity to the specific prompts and evidence truncation described in Appendix A. The paper acknowledges that confidence below 0.8 is mapped to neutral to reduce noise, but this does not establish that the judge's labels are correct or consistent. If the judge frequently misclassifies helpful reports as harmful or vice versa, the gating function could distort credit assignment in unpredictable ways, potentially explaining the observed improvements (or lack thereof) without actually testing the credit-assignment hypothesis. The paper's controlled comparison (Table 3) shows that LEAD-RL outperforms joint RL with shared credit, but this comparison does not isolate the effect of the rubric: the two conditions differ in both the credit-assignment rule and the presence of the judge's labels. A proper ablation would compare LEAD-RL against a version that uses the same judge labels but applies them randomly or with a fixed gate (e.g., always neutral). Without such an ablation, the observed gains could be due to the judge's specific (possibly biased) labeling rather than the credit-assignment logic itself. The paper's own limitation statements (e.g., 'the mechanism remains a hypothesis rather than a directly measured effect') underscore this concern.
The reader's weakest_assumption correctly identifies the reliance on an unvalidated rubric judge as a key concern. My analysis agrees and sharpens the point: the paper's central comparison does not disentangle the effect of the judge's accuracy from the credit-assignment mechanism itself. The proposed concrete test (random-label ablation) would directly address this. The paper's own acknowledgments of the hypothesis-like nature of the mechanism (b127, b181) support the need for such a test. I do not recommend changing the verdict from CONDITIONAL, because the concern is real but not fatal; the paper's results are still suggestive, and the proposed test is feasible. The agreement_with_reader is 'agree' because the reader's weakest_assumption and my load-bearing_attack converge on the same issue. The rationale explains why the concern is load-bearing and why the concrete test is appropriate.
The referee's report
If the results hold, the paper makes a meaningful contribution by addressing a specific credit-assignment problem in multi-agent delegation: the misalignment between team outcome and individual subagent report utility. The idea of using a rubric-based judge to gate the team advantage is novel and well-motivated. The experimental evaluation is thorough, covering multiple benchmarks and model families, with controlled ablations that isolate the effect of the credit-assignment mechanism. The paper also introduces useful efficiency metrics (critical-path cost, idealized speedup) that go beyond simple token counts. The main limitation is the reliance on an external judge (DeepSeek V4 Pro) for rubric labels without independent validation of its accuracy, which the authors acknowledge but do not fully address. Overall, the paper is a solid contribution to multi-agent RL for research agents.
Pith's reading of the paper (a machine's summary, not the authors')
Multi-agent deep-research systems use delegation: a main agent dispatches subagents to investigate subtasks in parallel, then synthesizes their reports. However, existing training methods either optimize only the main agent or give all agents the same team reward, which can reward misleading subagent reports and penalize useful ones. LEAD-RL addresses this by introducing a training-time rubric that classifies each subagent report as helpful, neutral, or harmful. The rubric gates the team advantage: helpful reports retain only positive advantages, harmful reports retain only negative ones, and neutral reports retain the original advantage. The main agent always receives the team advantage directly. This ensures that subagent credit is aligned with their actual contribution, not just the final outcome. The paper evaluates LEAD-RL on five benchmarks (BrowseComp, BrowseComp-ZH, GAIA, HLE, DeepSearchQA) using two model families (Ling-3-Flash and Qwen3.6-35B-A3B). On BrowseComp, LEAD-RL improves accuracy by 2.49 percentage points and reduces critical-path cost by 24.5% compared to joint training with shared team credit. Validation accuracy increases by 6.78 points at approximately constant critical-path cost. The trained models also compete with substantially larger systems, suggesting that better credit assignment can improve both research quality and computational efficiency.
Core claim. LEAD-RL improves answer quality and reduces critical-path cost in multi-agent deep research by using contribution-aware credit assignment that gates the team advantage based on a rubric assessment of each subagent report's research utility. On BrowseComp, it improves accuracy by 2.49 percentage points and reduces critical-path cost by 24.5% over joint training with shared team credit. During validation, accuracy increases by 6.78 points at approximately constant critical-path cost.
The premise it rests on. The rubric-based contribution judge (DeepSeek V4 Pro) can reliably classify subagent reports as helpful, neutral, or harmful based on the question, reference answer, brief, report, and evidence. The paper acknowledges that confidence below 0.8 is mapped to neutral to reduce noise, but the judge's accuracy is not independently validated. If the judge is unreliable, the credit assignment could be misdirected, potentially harming training. The paper does not provide an analysis of judge accuracy or inter-rater reliability.
Check this review
Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.
How often reviews like this one are right
Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.
How often objections from reviews like this one are right: not yet measured
Reviews by people
This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.
The paper
Abstract
Multi-agent deep-research systems use delegation to parallelize search and keep long tool traces out of the main agent's context. Yet delegation helps only when task allocation and subtask execution are coordinated: the main agent must issue useful subtasks, while subagents must return reliable evidence. Existing training either optimizes only the main agent or broadcasts the same team advantage to all agents, which can reward misleading subagent reports and penalize useful ones. We introduce LEAD-RL (Learning Effective Agent Delegation with Reinforcement Learning), a joint main-agent-subagent training framework with contribution-aware credit assignment. A training-time rubric evaluates the research utility of each subagent report and gates its inherited team advantage, while final-answer correctness remains the only team reward; no token, step, or speedup reward is used. Across five benchmarks and two model families, LEAD-RL improves answer quality. On BrowseComp, it improves accuracy by 2.49 percentage points and reduces critical-path cost by 24.5% over joint training with shared team credit. During validation, accuracy increases by 6.78 points at approximately constant critical-path cost. These results show that contribution-aware credit assignment improves the quality-efficiency trade-off of multi-agent deep research.
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.
abstract compiled · 2026-11-13
The authors' main claim, as the classifier picked it out 1.00
“These results show that contribution-aware credit assignment improves the quality-efficiency trade-off of multi-agent deep research.”
How the main claim is stated: asserted 1.00
What the abstract itself states, as probabilities: claims first 0.09 · states limitation 0.12 · human participants 0.03 · controlled comparison 0.94 · randomized assignment 0.02 · states quantitative result 0.99 · compares with prior results 0.96 · mentions code or data availability 0.03
The abstract, sentence by sentence
This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| Multi-agent deep-research systems use delegation to parallelize search and keep long tool traces out of the main agent's context. | background 1.00 | 0.06 | 0.02 | asserted 1.00 |
| Yet delegation helps only when task allocation and subtask execution are coordinated: the main agent must issue useful subtasks, while subagents must return reliable evidence. | background 0.72 | 0.07 | 0.06 | asserted 0.99 |
| Existing training either optimizes only the main agent or broadcasts the same team advantage to all agents, which can reward misleading subagent reports and penalize useful ones. | gap 0.93 | 0.05 | 0.03 | asserted 0.72 |
| We introduce LEAD-RL (Learning Effective Agent Delegation with Reinforcement Learning), a joint main-agent-subagent training framework with contribution-aware credit assignment. | contribution 0.93 | 0.90 | 0.09 | asserted 0.99 |
| A training-time rubric evaluates the research utility of each subagent report and gates its inherited team advantage, while final-answer correctness remains the only team reward; no token, step, or speedup reward is used. | method 0.96 | 0.89 | 0.08 | asserted 1.00 |
| Across five benchmarks and two model families, LEAD-RL improves answer quality. | result 0.99 | 0.94 | 0.27 | asserted 1.00 |
| On BrowseComp, it improves accuracy by 2.49 percentage points and reduces critical-path cost by 24.5% over joint training with shared team credit. | result 1.00 | 0.96 | 0.09 | asserted 1.00 |
| During validation, accuracy increases by 6.78 points at approximately constant critical-path cost. | result 1.00 | 0.93 | 0.07 | asserted 1.00 |
| These results show that contribution-aware credit assignment improves the quality-efficiency trade-off of multi-agent deep research. | interpretation 1.00 | 0.86 | 0.89 | asserted 1.00 |
What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.89
What was examined, and what was not
Sources read
- arXiv listing abstract v1
Not examined
- full text
- figures and tables
- references and citing sentences
- relations to other papers
Levels run
- P abstract
Not run
- S sections
- T body sentences
- C citations
- R cross-paper relations
How the labels were made
Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 6,892 input tokens.
Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON
Figures (3)
Figures are the authors', reproduced from the paper.
References (37)
- [1] Anthropic. Introducing claude sonnet 4.5, 2025.https://www.anthropic.com/news/claude-sonnet-4-5 on Pith
- [2] Center for AI Safety, Scale AI, and HLE Contributors Consortium. A benchmark of expert-level academic questions to assess ai capabilities.Nature, 649:1139–1146, 2026. doi: 10.1038/s41586-025-09962-4 on PithPith's reviewdoi:10.1038/s41586-025-09962-4
- [3] Zheng Chu, Xiao Wang, Jack Hong, Huiming Fan, Yuqi Huang, Yue Yang, Guohai Xu, Chenxiao Zhao, Cheng Xiang, Shengchao Hu, Dongdong Kuang, Ming Liu, Bing Qin, and Xing Yu. REDSearcher: A scalable and cost-efficient framework for long-horizon search agents.arXiv preprint arXiv:26... on PitharXiv
- [4] DeepSeek. Deepseek v4 preview release, 2026.https://api-docs.deepseek.com/news/news260424 on Pith
- [5] DeepSeek-AI. Deepseek-v3.2: Pushing the frontier of open large language models.arXiv preprint arXiv:2512.02556, 2025 on PithPith's reviewarXiv
- [6] Yuwen Du, Rui Ye, Shuo Tang, Xinyu Zhu, Yijun Lu, Yuzhu Cai, and Siheng Chen. OpenSeeker: Democratizing frontier search agents by fully open-sourcing training data. InConference on Language Modeling, 2026.https: //arxiv.org/abs/2603.15594 on PitharXiv
- [7] Jakob Foerster, Gregory Farquhar, Triantafyllos Afouras, Nantas Nardelli, and Shimon Whiteson. Counterfactual multi-agent policy gradients. InProceedings of the AAAI Conference on Artificial Intelligence, volume 32, 2018. https://ojs.aaai.org/index.php/AAAI/article/view/11794 on Pith
- [8] GLM Team. Glm-4.7: Advancing the coding capability, 2025.https://z.ai/blog/glm-4.7 on Pith
- [9] Nikita Gupta, Riju Chatterjee, Lukas Haas, Connie Tao, Andrew Wang, Chang Liu, Hidekazu Oiwa, Elena Gribovskaya, Jan Ackermann, John Blitzer, Sasha Goldshtein, and Dipanjan Das. Deepsearchqa: Bridging the comprehensiveness gap for deep research agents.arXiv preprint arXiv:2601... on PitharXiv
- [10] Ailin Huang, Ang Li, Aobo Kong, Bin Wang, Binxing Jiao, Bo Dong, Bojun Wang, Boyu Chen, Brian Li, Buyun Ma, et al. Step 3.5 flash: Open frontier-level intelligence with 11b active parameters.arXiv preprint arXiv:2602.10604, 2026 on PitharXiv
- [11] inclusionAI. Ling-3.0-flash. Hugging Face, 2026.https://huggingface.co/inclusionAI/Ling-3.0-flash on Pith
- [12] Bowen Jin, Hansi Zeng, Zhenrui Yue, Jinsung Yoon, Sercan O. Arik, Dong Wang, Hamed Zamani, and Jiawei Han. Search-R1: Training LLMs to reason and leverage search engines with reinforcement learning. InConference on Language Modeling, 2025.https://openreview.net/forum?id=Rwhi91ideu on Pith
- [13] Kimi Team. Kimi K2.5: Visual agentic intelligence.arXiv preprint arXiv:2602.02276, 2026.https://arxiv. org/abs/2602.02276 on PithPith's reviewarXiv
- [14] Kuan Li, Zhongwang Zhang, Huifeng Yin, Liwen Zhang, Litu Ou, Jialong Wu, Wenbiao Yin, Baixuan Li, Zheng- wei Tao, Xinyu Wang, Weizhou Shen, Junkai Zhang, Dingchu Zhang, Xixi Wu, Yong Jiang, Ming Yan, Pengjun Xie, Fei Huang, and Jingren Zhou. WebSailor: Navigating super-human r... on PithPith's reviewarXiv
- [15] Zhongyi Li, Wan Tian, Jinju Chen, Huiming Zhang, Yang Liu, Yikun Ban, and Fuzhen Zhuang. Counterfactual credit policy optimization for multi-agent collaboration.arXiv preprint arXiv:2603.21563, 2026.https://arxiv. org/abs/2603.21563 on PithPith's reviewarXiv
- [16] Muhan Lin, Shuyang Shi, Yue Guo, Vaishnav Tadiparthi, Behdad Chalaki, Ehsan Moradi Pari, Simon Stepputtis, Woojun Kim, Joseph Campbell, and Katia Sycara. Speaking the language of teamwork: LLM-guided credit assignment in multi-agent reinforcement learning.arXiv preprint arXiv:... on PithPith's reviewarXiv
- [17] Yijun Lu, Rui Ye, Yuwen Du, Jiajun Wang, Songhua Liu, and Siheng Chen. LongSeeker: Elastic context orchestration for long-horizon search agents.arXiv preprint arXiv:2605.05191, 2026.https://arxiv.org/abs/ 2605.05191 on PithPith's reviewarXiv
- [18] Meituan LongCat Team. Longcat-flash-thinking-2601 technical report.arXiv preprint arXiv:2601.16725, 2026 on PitharXiv
- [19] Grégoire Mialon, Clémentine Fourrier, Craig Swift, Thomas Wolf, Yann LeCun, and Thomas Scialom. Gaia: A benchmark for general ai assistants. InInternational Conference on 10 Learning Representations, 2024.https://proceedings.iclr.cc/paper_files/paper/2024/hash/ 25ae35b5b1738d8... on Pith
- [20] MiniMax-AI. Meet minimax-m2, 2025.https://github.com/MiniMax-AI/MiniMax-M2 on Pith
show 17 more
- [21] MiroMind Team. MiroThinker-1.7 & H1: Towards heavy-duty research agents via verification.arXiv preprint arXiv:2603.15726, 2026.https://arxiv.org/abs/2603.15726 on PitharXiv
- [22] Zhanfeng Mo, Xingxuan Li, Yuntao Chen, and Lidong Bing. Multi-agent tool-integrated policy optimization. arXiv preprint arXiv:2510.04678, 2025.https://arxiv.org/abs/2510.04678 on PitharXiv
- [23] Pu Ning, Quan Chen, Kun Tao, Xinyu Tang, Tianshu Wang, Qianggang Cao, Xinyu Kong, Zujie Wen, Zhiqiang Zhang, and Jun Zhou. SearchSwarm: Towards delegation intelligence in agentic LLMs for long-horizon deep research.arXiv preprint arXiv:2606.09730, 2026.https://arxiv.org/abs/26... on PithPith's reviewarXiv
- [24] OpenAI. Introducing gpt-5, 2025.https://openai.com/index/introducing-gpt-5/ on Pith
- [25] Qwen Team. Qwen3.6-35B-A3B. Hugging Face, 2026.https://huggingface.co/Qwen/Qwen3.6-35B-A3B on Pith
- [26] Qwen Team. Qwen3.8-2.4T-A95B. Hugging Face, 2026.https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B on Pith
- [27] Tabish Rashid, Mikayel Samvelyan, Christian Schroeder de Witt, Gregory Farquhar, Jakob Foerster, and Shimon Whiteson. Monotonicvaluefunctionfactorisationfordeepmulti-agentreinforcementlearning.Journal of Machine Learning Research, 21(178):1–51, 2020.https://jmlr.org/papers/v21... on Pith
- [28] Tongyi DeepResearch Team. Tongyi DeepResearch technical report.arXiv preprint arXiv:2510.24701, 2025. https://arxiv.org/abs/2510.24701 on PithPith's reviewarXiv
- [29] Junjie Wang, Zequn Xie, Dan Yang, Jie Feng, Yue Shen, Duolin Sun, Meixiu Long, Yihan Jiao, Zhehao Tan, Jian Wang, Peng Wei, and Jinjie Gu. WebClipper: Efficient evolution of web agents with graph-based trajectory pruning. InProceedings of the 64th Annual Meeting of the Associa... on Pithdoi:10.18653/v1/2026.acl-long.988.https://aclanthol
- [30] Jason Wei, Zhiqing Sun, Spencer Papay, Scott McKinney, Jeffrey Han, Isa Fulford, Hyung Won Chung, Alex Tachard Passos, William Fedus, and Amelia Glaese. Browsecomp: A simple yet challenging benchmark for browsing agents.arXiv preprint arXiv:2504.12516, 2025 on PithPith's reviewarXiv
- [31] Xixi Wu, Kuan Li, Yida Zhao, Liwen Zhang, Litu Ou, Huifeng Yin, Zhongwang Zhang, Xinmiao Yu, Dingchu Zhang, Yong Jiang, Pengjun Xie, Fei Huang, Minhao Cheng, Shuai Wang, Hong Cheng, and Jingren Zhou. ReSum: Unlocking long-horizon search intelligence via context summarization.a... on PitharXiv
- [32] Zequn Xie, Junjie Wang, Dan Yang, Jie Feng, Yue Shen, Jian Wang, and Jinjie Gu. SlimSearcher: Training efficiency-aware web agents via adaptive reward gating. InFindings of the Association for Computational Lin- guistics: EMNLP 2026, pages 10838–10853, Budapest, Hungary, Octob... on Pithdoi:10.18653/2026.findings-emnlp.517.https://aclanthology.org/2026.findings-emnlp.517/
- [33] Zelai Xu, Zhexuan Xu, Ruize Zhang, Chunyang Zhu, Shi Yu, Weilin Liu, Quanlu Zhang, Wenbo Ding, Chao Yu, and Yu Wang. WideSeek-R1: Exploring width scaling for broad information seeking via multi-agent reinforcement learning.arXiv preprint arXiv:2602.04634, 2026.https://arxiv.or... on PitharXiv
- [34] Rui Ye, Zhongwang Zhang, Kuan Li, Huifeng Yin, Zhengwei Tao, Yida Zhao, Liangcai Su, Liwen Zhang, Wenbiao Yin, Zile Qiao, Xinyu Wang, Pengjun Xie, Fei Huang, Jingren Zhou, Siheng Chen, and Yong Jiang. AgentFold: Long-horizon webagentswith proactive context folding. InInternati... on Pith
- [35] ChujieZheng, ShixuanLiu, MingzeLi, Xiong-HuiChen, BowenYu, ChangGao, KaiDang, YuqiongLiu, RuiMen, An Yang, Jingren Zhou, and Junyang Lin. Group sequence policy optimization.arXiv preprint arXiv:2507.18071, 2025.https://arxiv.org/abs/2507.18071 on PithPith's reviewarXiv
- [36] Yuxiang Zheng, Dayuan Fu, Xiangkun Hu, Xiaojie Cai, Lyumanshan Ye, Pengrui Lu, and Pengfei Liu. Deep- Researcher: Scaling deep research via reinforcement learning in real-world environments. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processi... on Pithdoi:10.18653/v1/2025.emnlp-main.22.https://aclanthology.org/2025.emnlp-main.22/
- [37] Peilin Zhou, Bruce Leon, Xiang Ying, Can Zhang, Yifan Shao, Qichen Ye, Dading Chong, Zhiling Jin, Chenxuan Xie, Meng Cao, et al. Browsecomp-zh: Benchmarking web browsing ability of large language models in chinese. arXiv preprint arXiv:2504.19314, 2025. 12 A Contribution Rubri... on PithPith's reviewarXiv
Challenges and discussion
Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.
No comments yet.
Sign in to comment