AISI Cyber TLO 10M: leaderboard

AISI cyber range: "The Last Ones", a 32-step corporate network attack simulation requiring reconnaissance, lateral movement, credential theft, reverse engineering, and data exfiltration. Avg steps completed at 10M token budget.

Metric: Avg Steps (/32). Source: arxiv.org. Status: years away from saturation. 10 models tracked.

Top models

#ModelScore
1Best Expert32
2GPT-5.510
3Claude Opus 4.69.8
4Claude Mythos Preview9.5
5GPT-5.48.8
6GPT-5.1 Codex8
7Claude Opus 4.57.6
8GPT-5.3 Codex7.2
9Claude Sonnet 4.56.1
10Claude 3.7 Sonnet5.8
11GPT-4o1.7

Interactive version: theaggregate.ai/benchmark?slug=aisi-cyber-tlo-10m · How It Works · Data refreshed daily, snapshot 2026-09-05.