Gemini 4 Argon hits 29/36 proven DFIR verdicts, best in 22-model security benchmark

DynamicWebPaige · x · 2026-10-16

Security Benchmarks' first Gemini 4 Argon evaluation uses an agentic SOC harness on the DFIR 0.6 collection:

• Basic investigations (36 cases): at High effort, Gemini 4 Argon returns 34/36 correct verdicts with accepted proof in 29 (80.6%) — the highest observed proof count across 22 models. GPT-6.1 Sol and GPT-6 Astra each reach 25, Claude Opus 5.5 only 12 (33.3%).
• Multi-host intrusions (12 cases): the proof lead vanishes — Gemini 4 Argon gets 4/12 with proof, behind GPT-6.1 Sol, GPT-6 Astra and Claude Opus 5.5 (5/12 each); Claude Opus 5.5 scores 100% correct verdicts.
• Models must classify commands as attack vs. routine administration, choose searches, inspect records, and cite supporting evidence.
• The author cautions that High effort doesn't imply equal internal compute across providers, and counts are from a single pass.

Original post →

More from Models

• Swapping RLM sub-calls for Jev: half the cost, 6x faster at 1M tokens — lateinteraction · 2026-10-16
• Stanford users say OpenAI cut ChatGPT Edu Pro usage 50% mid-contract — DJiafei · 2026-10-16
• MiniCPM-V 4.5 MoE reportedly hits 13+ tokens/s on iPhone with just 3GB memory — heyshrutimishra · 2026-10-16
• IQuest-Q1 writes a 629-line submarine game from one prompt in 2 minutes — Div_pradeep · 2026-10-16
• Trying Grok bot surfaces an unpaid Cursor balance, blocking account merging — haydendevs · 2026-10-16
• TensorFold 1.0.12 runs 120B vision model on 1-2 DGX Sparks, adds Mac image input — HankYeomans · 2026-10-16

Models channel →