SPADE Zero Shot - RAID Reviews 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
1SPADE Binoculars: Falcon-7B + Falcon-7B-InstructEvaluation run: frozen two-checkpoint detector67.1
2SPADE Binoculars: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector65.8
3SPADE Fast-DetectGPT: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector60.3
4falcon-7B (Harness: SPADE zero-shot NTS)56.2
5SPADE Binoculars: Qwen3-8B-Base + Qwen3-8BEvaluation run: frozen two-checkpoint detector34.9
6SPADE Fast-DetectGPT: Qwen3-8B-Base + Qwen3-8BEvaluation run: frozen two-checkpoint detector28.1
7SPADE Fast-DetectGPT: Falcon-7B + Falcon-7B-InstructEvaluation run: frozen two-checkpoint detector26
8Qwen3-8B-Base (Harness: SPADE zero-shot DNA-GPT)24
9falcon-7B (Harness: SPADE zero-shot DNA-GPT)21.9
10Llama 3.1 8B (Harness: SPADE Zero Shot NTS)21.2
11Llama 3.1 8B (Harness: SPADE Zero Shot DNA GPT)20.5
12Qwen3-8B-Base (Harness: SPADE zero-shot NTS)10.3
13Qwen3-8B-Base (Harness: SPADE zero-shot Likelihood)9.59
14Qwen3-8B-Base (Harness: SPADE zero-shot Entropy)9.59
15Llama 3.1 8B (Harness: SPADE Zero Shot Likelihood)9.59

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