Atlas

Benchmarks

← All benchmarks

CTF-Archive (CAISI 577-task set)

Agents · 2025-09-30

CAISI's 2025 CTF-Archive evaluation contains 577 challenges drawn from 79 competitions held from 2011 through 2025. Agents receive the original challenge description, relevant files, and an Inspect-based container environment, and scores report the percentage of flags retrieved.

Top models (higher is better)

ModelScore
GPT-550.6
gpt-oss-120b34.3
Opus 434.1
DeepSeek-V3.128.2
DeepSeek-R1-052826.5
DeepSeek-R18.5
Loading Atlas data…