A new benchmark and clean-room harness show frontier AI agents reach only 0.337 factual F1 when synthesizing conclusions from scientific evidence.
hub
Webexplorer: Exploreandevolvefortraininglong-horizonwebagents.arXivpreprint
17 Pith papers cite this work. Polarity classification is still indexing.
17
Pith papers citing it
hub tools
citation-role summary
background 2
baseline 1
citation-polarity summary
representative citing papers
SVI-Bench provides 35K hours of sports video with 9 tasks across four cognitive levels, revealing models drop from ~74% on action QA to 5% on agentic evidence integration.
The authors release ResearchMath-14k, the largest dataset of research-level math problems, and demonstrate that agent-filtered reasoning trajectories from open models improve fine-tuned Qwen3 models by 9.2 points on average.
Terminal-World is a skill-based synthesis pipeline that generates 5,723 training environments and produces Terminal-World-32B which outperforms baselines on Terminal-Bench 2.0 using only 1.2% of the data.
A verifiable offline Wikipedia tool environment plus iterative scaffold-to-ReAct self-distillation lets a 9B agent reach competitive BrowseComp/GAIA/HotpotQA scores without stronger-model distillation.
DecomposeR represents research plans as typed DAGs and uses two-stage planner-then-answerer RL to improve long-form research performance by 5.1-8.0 points over baselines.
GDCR assigns step-level rewards via distance to the answer node in a training-time ER graph and SAPO combines these with trajectory advantages for credit assignment in agentic search.
SR²AM achieves competitive Pass@1 accuracy on diverse tasks with 25.8-95.3% fewer reasoning tokens than much larger models by using self-regulated simulative planning trained via supervised learning and RL.
Argus coordinates a Navigator and multiple Searchers via an evidence graph for deep research, reporting average gains of 5.5 points with one Searcher and 12.7 points with eight parallel Searchers across eight benchmarks, reaching 86.2 on BrowseComp with 64 Searchers.
Retrievers trained on agent trajectories via the LRAT framework improve evidence recall, task success, and efficiency in agentic search benchmarks.
MiroThinker shows that scaling agent-environment interactions via reinforcement learning lets a 72B open-source model reach up to 81.9% on GAIA and approach commercial performance on research benchmarks.
Parallel WebBench reveals GRPO training raises web agent completion to 96% but leaves a large correctness gap from context-bound loops, premature termination, and synthesis collapse.
This survey categorizes agentic environments for LLMs by eight attributes and domains, introduces symbolic and neural synthesis paradigms with evaluation, and outlines four agent evolution pathways plus three environment evolution paradigms.
A 10-billion-parameter-activated Mixture-of-Experts model family is reported to match or approach larger frontier systems on several agentic coding, search, and office benchmarks.
SciResearcher is a new agentic data-construction framework that trains an 8B model via supervised fine-tuning and reinforcement learning to reach 19.46% on HLE-Bio/Chem-Gold and 13-15% gains on related biology and literature benchmarks.
Injecting 1% targeted synthetic data into GPT-2's pre-training substantially improves performance on 8 of 9 failing BLiMP grammatical paradigms, indicating data scarcity causes formal linguistic failures.
Retrieval-augmented generation reaches 0.66 weighted F1 for invalid bug report subclassification while agentic web search reaches 68.9% Judge LLM success for no-code fix generation.
citing papers explorer
-
Can AI Agents Synthesize Scientific Conclusions?
A new benchmark and clean-room harness show frontier AI agents reach only 0.337 factual F1 when synthesizing conclusions from scientific evidence.
-
SVI-Bench: A Dynamic Microworld for Strategic Video Intelligence
SVI-Bench provides 35K hours of sports video with 9 tasks across four cognitive levels, revealing models drop from ~74% on action QA to 5% on agentic evidence integration.
-
ResearchMath-14K: Scaling Research-Level Mathematics via Agents
The authors release ResearchMath-14k, the largest dataset of research-level math problems, and demonstrate that agent-filtered reasoning trajectories from open models improve fine-tuned Qwen3 models by 9.2 points on average.
-
Terminal-World: Scaling Terminal-Agent Environments via Agent Skills
Terminal-World is a skill-based synthesis pipeline that generates 5,723 training environments and produces Terminal-World-32B which outperforms baselines on Terminal-Bench 2.0 using only 1.2% of the data.
-
DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment
A verifiable offline Wikipedia tool environment plus iterative scaffold-to-ReAct self-distillation lets a 9B agent reach competitive BrowseComp/GAIA/HotpotQA scores without stronger-model distillation.
-
Planner-Centric Reinforcement Learning for Deep Research with Structure-Aware Reward
DecomposeR represents research plans as typed DAGs and uses two-stage planner-then-answerer RL to improve long-form research performance by 5.1-8.0 points over baselines.
-
Beyond Trajectory Rewards: Step-level Credit Assignment for Agentic Search via Graph Modeling
GDCR assigns step-level rewards via distance to the answer node in a training-time ER graph and SAPO combines these with trajectory advantages for credit assignment in agentic search.
-
Efficient Agentic Reasoning Through Self-Regulated Simulative Planning
SR²AM achieves competitive Pass@1 accuracy on diverse tasks with 25.8-95.3% fewer reasoning tokens than much larger models by using self-regulated simulative planning trained via supervised learning and RL.
-
Argus: Evidence Assembly for Scalable Deep Research Agents
Argus coordinates a Navigator and multiple Searchers via an evidence graph for deep research, reporting average gains of 5.5 points with one Searcher and 12.7 points with eight parallel Searchers across eight benchmarks, reaching 86.2 on BrowseComp with 64 Searchers.
-
Learning to Retrieve from Agent Trajectories
Retrievers trained on agent trajectories via the LRAT framework improve evidence recall, task success, and efficiency in agentic search benchmarks.
-
MiroThinker: Pushing the Performance Boundaries of Open-Source Research Agents via Model, Context, and Interactive Scaling
MiroThinker shows that scaling agent-environment interactions via reinforcement learning lets a 72B open-source model reach up to 81.9% on GAIA and approach commercial performance on research benchmarks.
-
When Web Agents Finish but Still Fail: Reproducible Triggers and Trace Diagnostics for Parallel Web Exploration
Parallel WebBench reveals GRPO training raises web agent completion to 96% but leaves a large correctness gap from context-bound loops, premature termination, and synthesis collapse.
-
Agentic Environment Engineering for Large Language Models: A Survey of Environment Modeling, Synthesis, Evaluation, and Application
This survey categorizes agentic environments for LLMs by eight attributes and domains, introduces symbolic and neural synthesis paradigms with evaluation, and outlines four agent evolution pathways plus three environment evolution paradigms.
-
The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence
A 10-billion-parameter-activated Mixture-of-Experts model family is reported to match or approach larger frontier systems on several agentic coding, search, and office benchmarks.
-
SciResearcher: Scaling Deep Research Agents for Frontier Scientific Reasoning
SciResearcher is a new agentic data-construction framework that trains an 8B model via supervised fine-tuning and reinforcement learning to reach 19.46% on HLE-Bio/Chem-Gold and 13-15% gains on related biology and literature benchmarks.
-
LiteResearcher: A Scalable Agentic RL Training Framework for Deep Research Agent
Injecting 1% targeted synthetic data into GPT-2's pre-training substantially improves performance on 8 of 9 failing BLiMP grammatical paradigms, indicating data scarcity causes formal linguistic failures.
-
Automated Root-Cause Subclassification and No-Code Fix Generation for Invalid Bug Reports
Retrieval-augmented generation reaches 0.66 weighted F1 for invalid bug report subclassification while agentic web search reaches 68.9% Judge LLM success for no-code fix generation.