Drift-Bench (State-Oriented, With Clarification) - Intention: leaderboard
Drift-Bench (Bao et al., University of Notre Dame, 2026) state-oriented set: 200 AgentBench operating-system and database tasks that at least two of three reference models solve with the clean instruction, rewritten so the instruction carries a changed intention (the stated goal differs from the one the user holds). The agent may call clarification tools (ask for a parameter, disambiguate, propose a solution, confirm a risk, report a blocker), answered by an LLM user simulator playing one of five decision-making personas. Success rate in percent, mean of three trials; higher is better.
Metric: Success rate (%; 200 AgentBench operating-system and database tasks with an intention fault (the stated goal changed), clarification tools answered by a persona user simulator, mean of 3 trials). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 235B A22B 2507 Instruct | 66.67 | #619 |
| 2 | Llama 4 Maverick | 59.33 | #1105 |
| 3 | GPT-5.2 | 55.9 | #172 |
| 4 | Gemini 2.5 Flash | 55.33 | #452 |
| 5 | DeepSeek V3.2 | 55.33 | #403 |
| 6 | GLM-4.7 | 52.17 | #365 |
| 7 | GPT-OSS-120B | 50.49 | #699 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=drift-bench-state-oriented-with-clarification-intention · How It Works · Data refreshed daily, snapshot 2026-11-12.