SPADE Zero Shot - EvoBench PubMed TPR at 5 Percent FPR: leaderboard

The SPADE paper compares six zero-shot AI-text detectors using three released language-model backbones. AUROC and true-positive rate at 5% false positives remain separate. Paired base/instruct checkpoints are retained for Binoculars and Fast-DetectGPT. Learned soft prompts and supervised detector weights are distinct artifacts. This board reports tPR at 5 Percent FPR (%). Higher is better.

Metric: TPR at 5 Percent FPR (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 18 models tracked. Updated 14 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen3-8B-Base (Harness: SPADE zero-shot Entropy)68.3
2Qwen3-8B-Base (Harness: SPADE zero-shot Likelihood)62.1
3SPADE Fast-DetectGPT: Falcon-7B + Falcon-7B-InstructEvaluation run: frozen two-checkpoint detector62.1
4SPADE Binoculars: Falcon-7B + Falcon-7B-InstructEvaluation run: frozen two-checkpoint detector59.6
5Qwen3-8B-Base (Harness: SPADE zero-shot DNA-GPT)48.7
6SPADE Binoculars: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector44.1
7SPADE Fast-DetectGPT: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector42.1
8falcon-7B (Harness: SPADE zero-shot DNA-GPT)41.3
9Llama 3.1 8B (Harness: SPADE Zero Shot Likelihood)39.5
10falcon-7B (Harness: SPADE zero-shot Likelihood)33.2
11Llama 3.1 8B (Harness: SPADE Zero Shot Entropy)32.4
12Llama 3.1 8B (Harness: SPADE Zero Shot DNA GPT)31.5
13SPADE Binoculars: Qwen3-8B-Base + Qwen3-8BEvaluation run: frozen two-checkpoint detector24
14SPADE Fast-DetectGPT: Qwen3-8B-Base + Qwen3-8BEvaluation run: frozen two-checkpoint detector22.3
15falcon-7B (Harness: SPADE zero-shot Entropy)15.9

Interactive version: theaggregate.ai/benchmark?slug=spade-zero-shot-evobench-pubmed-tpr-at-5-percent-fpr · How It Works · Data refreshed daily, snapshot 2026-11-16.