TACO: Tool-Augmented Credit Optimization for Agentic Tool Use
2026-06-29 • Multiagent Systems
Multiagent Systems
AI summaryⓘ
The authors developed a new training method called TACO to help AI models that answer questions about images by using extra code tools more effectively. Their method figures out exactly how helpful or harmful each tool use is by comparing the model’s answers with and without the tool, without needing a separate judge to evaluate. This way, the model learns to only use tools when they actually improve the answer. Testing showed that this approach improves accuracy and reduces unnecessary tool calls.
multimodal modelsvisual question answeringcode-tool agentsreinforcement learningadvantage functionself-supervised learningtool contributionoutcome rewardcredit assignmentsupervised fine-tuning
Authors
Mingkuan Feng, Jinyang Wu, Hao Gu, Fangrui Lv, Ruihan Jin, Chuyuan Zhang, Zhengqi Wen, Jianhua Tao
Abstract
Agentic multimodal models perform diverse operations on an image via code and reason over the returned view, an effective paradigm for fine-grained visual question answering. However, code operations can be useful, redundant, or misleading. Outcome-only rewards cannot precisely distinguish these cases, and existing process rewards either fail to attribute final correctness to individual tool calls, or require an external judge model. To address this, we introduce Tool-Augmented Credit Optimization (TACO), a GRPO variant for code-tool agents built on two coupled advantage channels. The first, Differential Answer-Probe Reward (DAPR), is a self-supervised, judge-free tool-contribution advantage that credits each tool call by its own effect on answering correctly. Probe tokens inserted into the model's reasoning elicit its predictions with and without the tool, and the difference in outcome reward is taken as the call's value: positive for a useful call, negative for a misleading one, and zero for one that changes nothing. This reuses the existing answer checker with no auxiliary judge, and, being a difference rather than an absolute probe score, is naturally robust to probe-hacking. The second is the outcome advantage from the final answer, distributed by Outcome-Gated Advantage Routing (OGAR): a parameter-free rule that, conditioned on the call's outcome, delivers this credit only to the responsible segments, suppressing wasted tool calls without any cost term. We train TACO through a two-stage SFT+RL pipeline. Extensive experiments across perception, reasoning, and general multimodal benchmarks show that it yields consistent accuracy gains and learns to invoke its tools only when they help.