Performance comparison of Procedural Graph against 7 baselines (Vanilla ReAct, MemoryBank, RAP, ExpeL, AutoGuide, AWM, KnowAgent) on τ-bench (customer service retail domain (115 test items, DB state match)) using Gemini 3.5 Flash. Procedural Graph achieves 44.35% compared to the best baseline score of 38.26% (margin of +6.09%).

model: Gemini 3.5 Flash · method: Vanilla ReAct (Yao et al., 2023b) · benchmark: τ-bench

model: Gemini 3.5 Flash · method: MemoryBank (Zhong et al., 2024) · benchmark: τ-bench

model: Gemini 3.5 Flash · method: RAP (Kagaya et al., 2024) · benchmark: τ-bench

model: Gemini 3.5 Flash · method: ExpeL (Zhao et al., 2024) · benchmark: τ-bench

model: Gemini 3.5 Flash · method: AutoGuide (Fu et al., 2024) · benchmark: τ-bench

model: Gemini 3.5 Flash · method: AWM (Wang et al., 2025b) · benchmark: τ-bench

model: Gemini 3.5 Flash · method: KnowAgent (Zhu et al., 2025) · benchmark: τ-bench

model: Gemini 3.5 Flash · method: Procedural Graph (Ours) · benchmark: τ-bench