MemEvoBench (Workflow Style): leaderboard
Metric: Attack success rate (%) in the third evaluation round on the 83 workflow-style cases (20 Agent-SafetyBench environments with noisy tool returns in the memory pool): share of responses that follow the misleading memory into the target risk behavior after two rounds of the agent's own responses were added to the memory pool, without user feedback, judged MISLED or SAFE by GPT-5.2, temperature 0, standard system prompt without safety guidance; lower is better. Source: arxiv.org. Saturation forecast: Around December 2027. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 69.2 |
| 2 | Claude 3.7 Sonnet | 82.7 |
| 3 | Qwen 3 235B A22B 2507 Instruct | 83.1 |
| 4 | Gemini 2.5 Pro | 84.3 |
| 5 | DeepSeek V3.2 | 84.3 |
| 6 | Qwen 3 Next 80B A3B Instruct | 84.3 |
| 7 | Qwen 3 32B | 89.2 |
| 8 | Llama 3.3 70B Instruct | 90.2 |
| 9 | GPT-4o | 91.6 |
Interactive version: theaggregate.ai/benchmark?slug=memevobench-workflow-style · How It Works · Data refreshed daily, snapshot 2026-10-07.