List your company

Tool-use Agent Performance on τ²-bench

72.4ASRMemMorph
4.59222.19639.857.404May 24, 2026

Evaluation results

MethodLinks
MemMorphAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.0572.4————————————
MemMorphAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0568.2————————————
MemMorphAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.0564.7————————————
MemMorphAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0560.5————————————
MemMorphAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0557.5————————————
MemMorphAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0556.1————————————
MemMorphAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.0555.9————————————
MemMorphAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.0553.8————————————
MemMorphAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0552.8————————————
GCGAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0552.5————————————
PoisonedRAGAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0551.5————————————
PoisonedRAGAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0551.2————————————
MemMorphAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0550.5————————————
PoisonedRAGAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.0546.3————————————
GCGAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.0545.5————————————
PoisonedRAGAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0545.5————————————
PoisonedRAGAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.0540.8————————————
PoisonedRAGAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.0538.6————————————
PoisonedRAGAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.0538.1————————————
PoisonedRAGAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0535.8————————————
PoisonedRAGAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0532.9————————————
GCGAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0532————————————
GCGAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0531.5————————————
ToolHijackerAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0530.6————————————
ToolCommanderAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0529.9————————————
GCGAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0528.7————————————
GCGAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.0527.9————————————
ToolCommanderAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0527.9————————————
PoisonedRAGAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0526.7————————————
ToolCommanderAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0525.7————————————
GCGAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0524.8————————————
ToolHijackerAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0524.3————————————
ToolCommanderAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.0523.4————————————
ToolCommanderAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.0521.8————————————
ToolHijackerAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0521————————————
GCGAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.0520.9————————————
GCGAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.0519.3————————————
ToolHijackerAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0518.6————————————
ToolHijackerAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.0518.6————————————
ToolCommanderAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0518.1————————————
ToolCommanderAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0516.7————————————
ToolHijackerAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.0516.3————————————
ToolHijackerAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0516.1————————————
ToolCommanderAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0515————————————
ToolHijackerAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0514.5————————————
ToolCommanderAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.0512.8————————————
ToolHijackerAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.0512.3————————————
GCGAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0511.2————————————
ToolHijackerAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.0510.9————————————
ToolCommanderAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.057.2————————————
A-MemModel=Qwen3-32B, Experience extraction model alignment=Aligned with inference model2026.04——————53.1238.7538.12————
A-MemModel=Kimi-K2-Instruct-0905, Experience extraction model alignment=Aligned with inference model2026.04——————76.2552.576.87————
A-MemModel=GLM-4.6, Experience extraction model alignment=Aligned with inference model2026.04——————76.887068.75————
AutoGuideModel=Qwen3-235B-A22B2026.01—32.221.114.949.0751.12———————
AutoGuideModel=Claude 3.7 Sonnet2026.01—45.336.633.360.6739.62———————
AWMModel=Qwen3-32B, Experience extraction model alignment=Aligned with inference model2026.04——————554038.12————
AWMModel=Qwen3-32B, Experience extraction model alignment=GLM-4.6 used for experience extraction2026.04——————57.541.2540.62————
AWMModel=Kimi-K2-Instruct-0905, Experience extraction model alignment=Aligned with inference model2026.04——————76.2553.7577.5————
AWMModel=Kimi-K2-Instruct-0905, Experience extraction model alignment=GLM-4.6 used for experience extraction2026.04——————76.2553.7577.5————
AWMModel=GLM-4.6, Experience extraction model alignment=Aligned with inference model2026.04——————77.571.2570.63————
ExpeLModel=Qwen3-32B, Experience extraction model alignment=Aligned with inference model2026.04——————56.2542.539.38————
ExpeLModel=Qwen3-32B, Experience extraction model alignment=GLM-4.6 used for experience extraction2026.04——————58.1243.7541.25————
ExpeLModel=Kimi-K2-Instruct-0905, Experience extraction model alignment=Aligned with inference model2026.04——————77.555.578.75————
ExpeLModel=Kimi-K2-Instruct-0905, Experience extraction model alignment=GLM-4.6 used for experience extraction2026.04——————77.556.2579.37————
ExpeLModel=GLM-4.6, Experience extraction model alignment=Aligned with inference model2026.04——————77.572.571.25————
IRMAMethod category=Agentic context-engineering2026.06—23.4———————9.653.1——
LedgerMethod category=Agentic context-engineering2026.06—27.2———————17.10——
LLM-onlyModel=Qwen3-235B-A22B2026.01—39.522.213.255.3743.84———————
LLM-onlyModel=Claude 3.7 Sonnet2026.01—46.233.927.263.2936.92———————
No MemoryModel=Qwen3-32B, Experience extraction model alignment=Aligned with inference model2026.04——————53.7538.7536.25————
No MemoryModel=Kimi-K2-Instruct-0905, Experience extraction model alignment=Aligned with inference model2026.04——————75.6251.2578.12————
No MemoryModel=GLM-4.6, Experience extraction model alignment=Aligned with inference model2026.04——————76.257070.63————
Off-policy Distill.Backbone=Qwen3-30B-A3B-Thinking-2507, Task Setup=Single-Task2026.06———————60.592——41.774.7
Off-policy Distill.Backbone=Qwen3-30B-A3B-Thinking-2507, Task Setup=Multi-Task2026.06———————6188——41.565.8
Off. + On.Backbone=Qwen3-30B-A3B-Thinking-2507, Task Setup=Multi-Task2026.06———————6292.2——43.874.7
On-policy Distill.Backbone=Qwen3-30B-A3B-Thinking-2507, Task Setup=Multi-Task2026.06———————61.188.8——4172.2
Pre-RLBackbone=Qwen3-30B-A3B-Thinking-2507, Task Setup=Single-Task2026.06———————52.977.4——30.951.1
ReActModel=Qwen3-235B-A22B2026.01—34.822.218.442.1757.5———————
ReActModel=Claude 3.7 Sonnet2026.01—50.635.927.863.5736.59———————
ReflexionModel=Qwen3-235B-A22B2026.01—34.521.616.754.5645.66———————
ReflexionModel=Claude 3.7 Sonnet2026.01—53.543.337.765.434.65———————
REMEMBERERModel=Qwen3-235B-A22B2026.01—31.916.49.747.9252.05———————
REMEMBERERModel=Claude 3.7 Sonnet2026.01—52.342.436.866.2933.92———————
RLBackbone=Qwen3-30B-A3B-Thinking-2507, Task Setup=Single-Task2026.06———————61.692.8——42.676.4
SkillXModel=Qwen3-32B, Experience extraction model alignment=GLM-4.6 used for experience extraction2026.04——————66.8747.543.75————
SkillXModel=Kimi-K2-Instruct-0905, Experience extraction model alignment=GLM-4.6 used for experience extraction2026.04——————78.1258.7582.5————
SkillXModel=GLM-4.6, Experience extraction model alignment=Aligned with inference model2026.04——————82.576.2571.88————
WISE-FlowModel=Qwen3-235B-A22B2026.01—39.525.219.363.4336.65———————
WISE-FlowModel=Claude 3.7 Sonnet2026.01—56.446.842.169.3630.75———————