Zvi digs into Gemini 4 Argon: 14-page model card vs 217-page safety report
Don't Worry About the Vase (Zvi) · rss · 2026-10-26
Zvi's deep read of Gemini 4 Argon's model card and 217-page safety report: much stronger than Flash 3.8; cannot rule out crossing Cyber L1, CBRN L1, and Stealth TCL thresholds; world-class prompt injection resistance, hallucinations down two-thirds; higher eval awareness; CoT readable but defeatable at a performance cost; plus criticism of selective benchmark contrasts in the safety card.
More from Models
- DeepSeek vs Haiku pricing debunk: accounting for peak rates and token use, 1017 is 2-2.5x cheaper — teortaxesTex · 2026-10-28
- ChatGPT and Gemini refuse to draft a complaint against a pedophile, user reports — chinchinlover-419 · 2026-10-28
- GPT-6.1 Terra priced at $1/$5 per M tokens, but the '9x cheaper per task' math doesn't hold — 2C_ornot2C · 2026-10-28
- FrontierCode 2 leaderboard sparks drama over missing Chinese models — teortaxesTex · 2026-10-28
- Evil 27B Too Risky to Run Autonomous? User Fears Wiped Drive Over a Game — teortaxesTex · 2026-10-28
- DeepSeek V5 rumor debunked: code strings were just September assert tests — teortaxesTex · 2026-10-28