Tool-use on Task-Bench
58.2Task Completion RateReasoningbank
Evaluation results
| Method | Links | |
|---|---|---|
| ReasoningbankLLM Backbone=Qwen3-32B2026.02 | 58.2 | |
| No-MemoryLLM Backbone=Qwen3-32B2026.02 | 56.8 | |
| Reasoningbank+promptLLM Backbone=Qwen3-32B2026.02 | 56.4 | |
| DCLLM Backbone=Qwen3-32B2026.02 | 54 | |
| Reasoningbank+GuardLLM Backbone=Qwen3-32B2026.02 | 52.2 | |
| MementoLLM Backbone=Qwen3-32B2026.02 | 51.8 | |
| TAMELLM Backbone=Qwen3-32B2026.02 | 51.8 | |
| No-MemoryLLM Backbone=GPT-5.22026.02 | 48.6 | |
| TAMELLM Backbone=GPT-5.22026.02 | 46.8 | |
| Reasoningbank+GuardLLM Backbone=GPT-5.22026.02 | 45.4 | |
| ReasoningbankLLM Backbone=GPT-5.22026.02 | 44.8 | |
| Reasoningbank+promptLLM Backbone=GPT-5.22026.02 | 44.2 | |
| DCLLM Backbone=GPT-5.22026.02 | 43.8 | |
| MementoLLM Backbone=GPT-5.22026.02 | 40 |