Skip to content
The Aggregate

Unified LLM rankings, daily

Models What's New Trends Benchmarks Skill Maps Task Explorer Builder
How It Works About Guesswork Metrics
The Aggregate
Loading data...

When the Manual Lies: leaderboard

Metric: Attack Success Rate (ASR) (self-reported). Source: benchmarklist.com. 8 models tracked.

Top models

#ModelScore
1Qwen 3 32B98
2GPT-4o100
3Claude 3.7 Sonnet100
4Claude Sonnet 4100
5DeepSeek R1100
6QwQ-32B100
7DeepSeek V3100

Interactive version: theaggregate.ai/benchmark?slug=when-the-manual-lies · How It Works · Data refreshed daily, snapshot 2026-09-05.

Built by Mikhail Doroshenko — AI researcher, co-author of Humanity’s Last Exam. Independent project; no affiliation with any AI lab.
A

Your Aggregate

Sign in to keep your work

Save custom rankings and get 50 free task analyses each day.

Continue with Google Continue with GitHub

No password to remember. Your public profile identifies saved work.

Your Aggregate

Account

  • Saved custom aggregate views
  • 50 free task analyses per day
Sign out