# CARE-PO: Controllable Tool-Call Efficiency for LLM Agents

DevFeed: [CARE-PO: Controllable Tool-Call Efficiency for LLM Agents](<https://devfeed.tech/articles/care-po-controllable-tool-call-efficiency-for-llm-agents-125516.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.12125>)

Author: Xiaoyang Ye, Xiao Yao, Hairui Zhu, Yiying Yang, Bei Gan, Jinliang Lu, Tengjin Weng, Yuan Yao, Wenhao Jiang

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [grpo](<https://devfeed.tech/topics/grpo.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [gaia](<https://devfeed.tech/topics/gaia.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [budget](<https://devfeed.tech/tags/budget.md>), [call](<https://devfeed.tech/tags/call.md>), [capability](<https://devfeed.tech/tags/capability.md>), [charging](<https://devfeed.tech/tags/charging.md>), [completion](<https://devfeed.tech/tags/completion.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [diagnosis](<https://devfeed.tech/tags/diagnosis.md>), [ease](<https://devfeed.tech/tags/ease.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [published](<https://devfeed.tech/tags/published.md>), [tool](<https://devfeed.tech/tags/tool.md>)

## AI overview

CARE-PO aims to reduce unnecessary external tool calls while preserving task success. It prices calls against a fixed reference budget based on task difficulty, adds a successful training trajectory to each online GRPO rollout group, and charges for evidence a trajectory never gathers to discourage policies from collapsing to zero calls. A prompt effort tag shapes the cost curve, allowing one training run to produce different efficiency-capability operating points. On GAIA, five of six tested points across three model sizes improved both success rate and call count over their base models. On four held-out multi-hop benchmarks, results depended on size: at 27B and 4B, success was maintained or improved with fewer calls; at 9B, both points traded lower success for fewer calls. Call counts stayed above zero during training.

## Source excerpt

arXiv:2611.12125v1 Announce Type: new Abstract: External tool calls are a major running cost of an LLM agent, while its training objectives align almost exclusively with task completion. In a step-level diagnosis of Qwen3.5 trajectories at two model sizes on GAIA, a long-horizon tool-use benchmark, we find that most contain removable calls, at a similar rate at both sizes. Some methods add an efficiency reward to the objective to reduce the call count, but introduce problems of their own: the anchor they price calls against is computed from the policy's own rollouts (e.g., the group's shortest correct rollout) and so moves with the policy; training can drift toward a collapsed policy that issues no tool call at all, a degeneracy seen in published reports and in our 27B baseline; and a training run delivers a single operating point, so the user cannot trade efficiency against capability. We present CARE-PO (Controllable Anchored Relative Efficiency Policy Optimization) to mitigate these problems. The efficiency reward prices calls against a reference budget anchored to the task's own difficulty and fixed before training. Injecting a successful trajectory from the training data into each online GRPO rollout group and charging for evidence a trajectory never gathers ease the drift toward a collapsed call count. Shaping the cost curve by an effort tag in the prompt yields an efficiency and a capability operating point from a single training run. On GAIA, five of the six operating points across three model sizes improve on their base model in success rate and call count at once. On four held-out multi-hop benchmarks the result is size-dependent: at 27B and 4B success rate is held or raised at a lower call count, while at 9B both operating points trade success rate for fewer calls. The call count remains well above zero throughout training at all three sizes.