Graph Reasoning on GraphArena
100Average AccAgentSquare
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AgentSquareBase LLM=gpt-5.4-nano, Category=Tool-Augmented Agents, Tool Access=Fixed Toolbox2026.05 | 100 | — | — | |
| MA-GTSBase LLM=gpt-5.4-nano, Category=Tool-Augmented Agents, Tool Access=Fixed Toolbox2026.05 | 100 | — | — | |
| w/o SCA RoutingBase LLM=gpt-5.4-nano, Ablation=w/o SCA Routing2026.05 | 99.5 | — | — | |
| w/o Tool GrowthBase LLM=gpt-5.4-nano, Ablation=w/o Tool Growth2026.05 | 99.5 | — | — | |
| w/o Protocol-Reliability ObjBase LLM=gpt-5.4-nano, Ablation=w/o Protocol-Reliability Obj2026.05 | 99.5 | — | — | |
| EGL-SCA (Full)Base LLM=gpt-5.4-nano2026.05 | 99.5 | — | — | |
| w/o Instruction Evol.Base LLM=gpt-5.4-nano, Ablation=w/o Instruction Evol.2026.05 | 91 | — | — | |
| FixedSolverBase LLM=gpt-5.4-nano, Category=Tool-Augmented Agents, Tool Access=Fixed Toolbox2026.05 | 67 | — | — | |
| ReActBase LLM=gpt-5.4-nano, Category=Tool-Augmented Agents, Tool Access=Fixed Toolbox2026.05 | 65.5 | — | — | |
| NPG-Muse-8BParameters=8B2025.08 | 54.4 | 63.2 | 32.9 | |
| GEPABase LLM=gpt-5.4-nano, Category=Prompting & Self-Improvement, Tool Access=No Tools2026.05 | 47.5 | — | — | |
| Chain-of-Thought (CoT)Base LLM=gpt-5.4-nano, Category=Prompting & Self-Improvement, Tool Access=No Tools2026.05 | 42 | — | — | |
| Qwen2.5-14B-Instruct-1MBackbone=Qwen2.5, Parameters=14B, Variant=Instruct-1M2025.08 | 40.9 | 48.1 | 15.8 | |
| NPG-Muse-7BParameters=7B2025.08 | 36.2 | 47.7 | 10.7 | |
| w/o Tool UseBase LLM=gpt-5.4-nano, Ablation=w/o Tool Use2026.05 | 36 | — | — | |
| Qwen2.5-7B-Ins-1MBackbone=Qwen2.5, Parameters=7B, Variant=Instruct-1M2025.08 | 32.9 | 37.2 | 15.2 | |
| Qwen3-14B-BaseBackbone=Qwen3, Parameters=14B2025.08 | 32.6 | 31.6 | 14.8 | |
| Qwen3-8B-BaseBackbone=Qwen3, Parameters=8B2025.08 | 25.2 | 20.8 | 9.5 | |
| ExpeLBase LLM=gpt-5.4-nano, Category=Prompting & Self-Improvement, Tool Access=No Tools2026.05 | 18 | — | — | |
| Direct PromptingBase LLM=gpt-5.4-nano, Category=Prompting & Self-Improvement, Tool Access=No Tools2026.05 | 16 | — | — | |
| Few-ShotBase LLM=gpt-5.4-nano, Category=Prompting & Self-Improvement, Tool Access=No Tools2026.05 | 8 | — | — | |
| ReflexionBase LLM=gpt-5.4-nano, Category=Prompting & Self-Improvement, Tool Access=No Tools2026.05 | 4.5 | — | — |