DiagFlowBench - Correct Abstention: leaderboard
Metric: Correct abstention (%): share of injected off-procedure operator turns (coverage gaps, undocumented malfunctions, unrelated questions) on which the model declines instead of mapping to a real but inadequate step or fabricating one, classified by a claude-haiku-4-5 judge, on 1,676 multi-turn maintenance conversations over 50 industrial diagnostic flowcharts from a consumer manufacturer, procedure graph given as JSON in the system prompt, temperature 0, via OpenRouter; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3.3 70B | 81.3 |
| 2 | Nemotron 3 Super | 73.9 |
| 3 | Gemini 2.5 Flash | 72 |
| 4 | Llama 4 Maverick | 65.7 |
| 5 | Qwen 3 235B A22B (Thinking) | 64.9 |
| 6 | GPT-4o Mini | 58.3 |
| 7 | GPT-OSS-120B | 54.7 |
| 8 | Llama 4 Scout | 39.4 |
| 9 | Qwen 3 30B A3B (Thinking) | 27.4 |
Interactive version: theaggregate.ai/benchmark?slug=diagflowbench-correct-abstention · How It Works · Data refreshed daily, snapshot 2026-09-29.