Coding · sub-raceDominant lead
Cybersecurity race
Capture-the-flag security challenges solved without hints.
- Leader
- Opus 4.6
- Margin
- 33.0 pts
- 55.0% ahead
- Weeks at #1
- 38
- in a row
- Changes
- 1
- in 2026
Warming up the track…
Oct 20, 2026Opus 4.6 leads
2025Apr ’25Jul ’25Oct ’252026Apr ’26Jul ’26Oct ’26
Score history
Top 4 today over the leader's frontier · click to jump to a week
#1 frontierOpus 4.6Sonnet 4.5Grok 4.1o3-mini
Reigns at #1
Longest: Opus 4.6, 38 weeks (from Feb 2026).
Sibling races
More coding races
Why this race matters
Offensive-security skill matters for defenders and for risk assessments of frontier models.
How it is scored
Cybench unguided solve rate from Epoch AI's Benchmarking Hub. As measured, placed at release date. Few current models are covered.
Unit: percentHigher is betterPhoto finish under 0.5 ptsDominant over 3.0 pts
Data provenance
Epoch AI, 'Capabilities & benchmarking', CC BY 4.0.
Cybench · Epoch AI
As measured: one result per model, placed at its release date.
10 tracked models with a result (22 entries on the source board) · newest data February 5, 2026.
Partial coverage: the source does not report every tracked model, so treat the order as indicative.