Report
Gemini 4 Argon gets 29 of 36 basic verdicts with accepted proof in a preliminary benchmark
Security Benchmarks reports Gemini reached 4 of 12 multi-host intrusion verdicts with proof, trailing three rivals.
TLDR
Security Benchmarks reports that Gemini 4 Argon, at High effort, delivered accepted proof for 29 of 36 basic investigation verdicts—the highest count in its 22-model DFIR 0.6 collection. On multi-host intrusions, it got proof for 4 of 12 verdicts, versus 5 of 12 for GPT-6.1 Sol, GPT-6 Astra and Claude Opus 5.5. The results come from one pass per model and, the publisher notes, do not establish statistical superiority.
Combined views
7
1 Source, first seen ago
