SPADE Zero Shot - RAID Recipes AUROC: 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 aUROC (%). Higher is better.

Metric: AUROC (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 18 models tracked. Updated 14 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1falcon-7B (Harness: SPADE zero-shot NTS)87.9
2SPADE Binoculars: Falcon-7B + Falcon-7B-InstructEvaluation run: frozen two-checkpoint detector87.3
3SPADE Binoculars: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector85.2
4SPADE Fast-DetectGPT: Falcon-7B + Falcon-7B-InstructEvaluation run: frozen two-checkpoint detector83.6
5SPADE Fast-DetectGPT: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector82.3
6falcon-7B (Harness: SPADE zero-shot DNA-GPT)78.2
7Qwen3-8B-Base (Harness: SPADE zero-shot Entropy)77.7
8Qwen3-8B-Base (Harness: SPADE zero-shot Likelihood)77.1
9falcon-7B (Harness: SPADE zero-shot Likelihood)74.8
10Qwen3-8B-Base (Harness: SPADE zero-shot NTS)74.4
11Llama 3.1 8B (Harness: SPADE Zero Shot Entropy)73
12Llama 3.1 8B (Harness: SPADE Zero Shot Likelihood)72.9
13Llama 3.1 8B (Harness: SPADE Zero Shot NTS)70.2
14falcon-7B (Harness: SPADE zero-shot Entropy)68.8
15Qwen3-8B-Base (Harness: SPADE zero-shot DNA-GPT)67.4

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