SPADE Zero Shot - RAID Recipes 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 detector75
2SPADE Binoculars: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector70.9
3falcon-7B (Harness: SPADE zero-shot NTS)67.4
4SPADE Fast-DetectGPT: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector56
5falcon-7B (Harness: SPADE zero-shot DNA-GPT)54.6
6SPADE Fast-DetectGPT: Falcon-7B + Falcon-7B-InstructEvaluation run: frozen two-checkpoint detector51.8
7Llama 3.1 8B (Harness: SPADE Zero Shot NTS)30.5
8falcon-7B (Harness: SPADE zero-shot Likelihood)24.8
9SPADE Binoculars: Qwen3-8B-Base + Qwen3-8BEvaluation run: frozen two-checkpoint detector24.1
10Qwen3-8B-Base (Harness: SPADE zero-shot NTS)21.3
11Qwen3-8B-Base (Harness: SPADE zero-shot Entropy)16.3
12SPADE Fast-DetectGPT: Qwen3-8B-Base + Qwen3-8BEvaluation run: frozen two-checkpoint detector15.6
13Qwen3-8B-Base (Harness: SPADE zero-shot Likelihood)14.2
14Qwen3-8B-Base (Harness: SPADE zero-shot DNA-GPT)14.2
15Llama 3.1 8B (Harness: SPADE Zero Shot DNA GPT)12.1

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