Large language models (LLMs) are increasingly extended into deep search agents that solve complex questions through multi-step interaction with external search and browsing tools. However, existing agents often incur substantial computational and interaction costs, generating lengthy trajectories that contain redundant queries, inefficient exploration, and irrelevant observations. Existing efficiency-oriented methods usually encourage agents to use tools less frequently, but treating all tool interactions uniformly may also suppress steps that gather necessary evidence. In this paper, we propose CRISP, a framework for training efficient deep search agents through critical step perception. Unlike prior efficiency methods that uniformly penalize tool use, CRISP distinguishes interactions that gather necessary evidence from redundant ones and shapes the training reward to preserve the former while pruning the latter, improving efficiency without sacrificing the evidence needed for correct answers. Specifically, CRISP first constructs critical-step labels with Backward Evidence Induction: starting from the final answer, a strong model traverses a completed search trajectory backward and judges whether each tool-interaction step provides or preserves evidence for the final answer. We then distill these step-wise judgments into a smaller critical-step recognizer, enabling full-trajectory analysis in a single pass. During policy optimization, an efficiency-aware reward is applied only to successful rollouts. Experiments on BrowseComp and HLE-Verified show that CRISP maintains competitive final-answer accuracy while reducing average interaction turns by 15.1% and 33.2%, respectively, demonstrating substantial improvements in interaction efficiency.
As Large Language Models (LLMs) evolve into autonomous agents, traditional static evaluation fails to capture multi-step decision-making. We introduce AgenticAI-Supervisor, an API and UI-driven RL Gym environment that decouples environment creation from scalable execution. By moving to verifiable execution outcomes, the platform generates high-fidelity traces and applies multi-dimensional reward shaping. Critically, our framework mitigates reward hacking through rigorous internal state validation and testing. This work provides a first look at our platform's core capabilities through a Customer Support Agent case study demonstrating a consistent closed-loop feedback for model optimization. Future work will focus on advanced features such as Computer Use, Tool Use, automated "stumping", and edge-case generation.
Training multi-turn evidence-reading agents with outcome-only reinforcement learning is unstable because intermediate turns receive little direct credit. In HotpotQA experiments with Qwen2.5-3B-Instruct, GRPO initially improves (standard F1 0.430) but subsequently collapses to 100% format-violating outputs. Training-log diagnosis reveals a zero-advantage lock-in mechanism: all sampled trajectories receive the minimum format penalty (-2.0), group-relative advantages vanish, and the policy-gradient loss becomes zero--an optimization deadlock. We propose a variance-injection strategy: by assigning per-turn rewards to intermediate evidence-reading turns, we prevent the group reward distribution from collapsing to a single value--preserving the variation that GRPO's group-relative advantage requires. Contextual Information-Gain Policy Optimization (CIGPO) implements this strategy using the marginal increase in the frozen reference model's log-likelihood of the ground-truth answer as the per-turn signal. With separate normalization of IG and F1 rewards and an IG-weight curriculum, CIGPO reaches a standard F1 of 0.518 on HotpotQA at the 3B scale (from 0.252 base; +105%), compared with 0.430 for the best GRPO checkpoint and 0.000 for the final GRPO checkpoint. CIGPO maintains meaningful reward variance and avoids zero-advantage lock-in throughout training. These results identify reward-variance collapse as a concrete failure mode of outcome-only GRPO and show that turn-level IG rewards can prevent it in this HotpotQA setting.
We measure how well current large language models coordinate as multiple agents sharing a common resource, using the dining philosophers problem as a clean test bed. Across 630 episodes spanning seven models and three philosopher counts, four frontier closed-source systems reach mean reward 0.45 to 0.87 and Mistral-Small 24B reaches 0.83 to 0.99, while Qwen3-14B reaches 0.13 to 0.35. We then ask whether group relative policy optimization (GRPO) on rollouts from the task itself can close the gap and find that it cannot: a Welch's t-test on per-episode reward at five philosophers gives p = 0.66 and a Hedges' g of -0.11, with no statistically significant change at ten or fifteen philosophers either. Two further observations qualify the result. The training reward of both 8B and 14B runs peaked at step nine and then declined, so the default saved checkpoint at step 15 is strictly worse than several earlier ones. The four-term reward we use admits a degenerate maximum at zero actions, which DeepSeek-R1-Distill-Qwen-7B and Mistral-Small 24B at five philosophers both inhabit, with mean reward 1.0 and 0.83 respectively at zero meals. The bottleneck for an open-weight 14B model on multi-agent coordination is not training compute but training methodology: reward shaping that does not collapse to a no-action maximum, checkpoint discipline that does not depend on the final step, and curriculum across problem scales.
Deep research agents have demonstrated remarkable capabilities in complex information-seeking tasks, yet this power comes at a steep computational cost. Driven by accuracy-focused training paradigms, current models adopt brute-force strategies characterized by blind tool dependency and performative reasoning-generating long, redundant trajectories that are far from necessary for resolving these tasks, leading to wasteful tool calls and excessive token consumption. To overcome this efficiency trap, we propose SlimSearcher, a principled framework that pushes the Pareto frontier between accuracy and computational cost across both Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL). In the SFT stage, SlimSearcher employs Pareto-efficient filtration to distill trajectories that are both successful and economical, guiding the model toward inherently efficiency-aware search behaviors. During RL, we introduce Adaptive Reward Gating, a dynamic reward-shaping mechanism that evaluates relative tool and token efficiency within a sampled cohort. By cascading these adaptive efficiency metrics with a strict correctness gate, our approach effectively avoids the brevity bias associated with absolute penalties and mitigates reward hacking. Extensive experiments on long-horizon benchmarks, including GAIA, BrowseComp, and XBenchDeepSearch, demonstrate that SlimSearcher reduces average tool-call rounds by 17%-58% while maintaining or improving accuracy.
Agentic reinforcement learning can induce tool abuse, where models overuse external tools even for queries solvable by internal reasoning. Existing approaches mitigate this issue with uniform tool-use penalties or hard limits, which reduce tool frequency but may also suppress useful tool-assisted exploration. We propose EAPO, an Efficient Agentic Policy Optimization framework that learns selective tool use. EAPO introduces tool-free trajectories into each rollout group, applies difficulty-aware reward shaping to penalize redundant tool calls mainly on easier queries, and uses confidence-aware token reweighting to improve policy learning. Across nine mathematical and knowledge-intensive reasoning benchmarks, EAPO consistently improves the accuracy efficiency trade-off on Qwen2.5-3B, Qwen2.5-7B, and Llama3.1-8B. Compared with GRPO, EAPO improves average performance by 10.45%, 7.27%, and 9.69%, while reducing average tool calls by 18.33%, 18.33%, and 24.59%, respectively. These results show that agents can learn when not to use tools without compromising tool-integrated reasoning.