PhantomBench - Entities (Existence): leaderboard

Metric: Hallucination rate (%): share of answers that do not abstain on the Phantom-E subset (1,200 generated non-existent entities: events, holidays, elections, disasters, creative works and places), verified absent from the 2.3-trillion-token Dolma corpus, when the prompt asks whether the concept exists; a Gemini 2.5 Flash judge labels abstention; reasoning-model rates count only generations that reached a final answer; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.

Top models

#ModelScore
1Qwen 3 14B3.5
2GPT-OSS-20B (Medium)3.93
3Qwen 3 8B4
4Qwen 2.5 7B Instruct4.08
5GPT-OSS-20B (Low)4.68
6Gemma 2 9B (IT)7.16
7Llama 3.1 8B Instruct7.41
8Qwen 3 4B7.5
9Qwen 3 32B8
10Qwen 3 1.7B13.59
11DeepSeek R1 Distill Qwen 32B15.2
12Mistral 7B Instruct (v0.3)17.09
13Gemini 2.5 Flash22.25
14Gemini 2.5 Pro29.33
15Gemma 3 12B (IT)47.5

Interactive version: theaggregate.ai/benchmark?slug=phantombench-entities-existence · How It Works · Data refreshed daily, snapshot 2026-09-29.