SPADE Zero Shot - EvoBench Overall 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: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector56.9
2SPADE Fast-DetectGPT: Llama 3.1 8B + Llama 3.1 8B InstructEvaluation run: frozen two-checkpoint detector55.6
3SPADE Binoculars: Falcon-7B + Falcon-7B-InstructEvaluation run: frozen two-checkpoint detector52.3
4SPADE Fast-DetectGPT: Falcon-7B + Falcon-7B-InstructEvaluation run: frozen two-checkpoint detector51.3
5Qwen3-8B-Base (Harness: SPADE zero-shot Likelihood)49.3
6Qwen3-8B-Base (Harness: SPADE zero-shot Entropy)42.8
7Qwen3-8B-Base (Harness: SPADE zero-shot DNA-GPT)41.4
8Llama 3.1 8B (Harness: SPADE Zero Shot Likelihood)36.9
9falcon-7B (Harness: SPADE zero-shot DNA-GPT)35.1
10SPADE Binoculars: Qwen3-8B-Base + Qwen3-8BEvaluation run: frozen two-checkpoint detector33.4
11Llama 3.1 8B (Harness: SPADE Zero Shot DNA GPT)32.2
12SPADE Fast-DetectGPT: Qwen3-8B-Base + Qwen3-8BEvaluation run: frozen two-checkpoint detector30.1
13falcon-7B (Harness: SPADE zero-shot NTS)29.3
14falcon-7B (Harness: SPADE zero-shot Likelihood)29.1
15Llama 3.1 8B (Harness: SPADE Zero Shot Entropy)26.7

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