Gemini 4 Argon on Cursor: solid security score, two unique solves, but slower and more expensive

Google’s Gemini 4 Argon, evaluated with the Cursor coding harness on security-sensitive software tasks, tied Fable 5 for the top Cursor SecPass score of 29% while achieving a 78.8% FuncPass rate, but ranked sixth across all tested harnesses behind several faster and less costly competitors. Argon independently produced two security fixes that no other evaluated combination solved, addressing a path traversal flaw in SABnzbd and credential exposure in OpenStack Nova, with both requiring extensive code exploration and more than 20 minutes of execution. However, its median task time of 14 minutes and total test cost of $765 made it the slowest and most expensive recent frontier combination, contrasting with models such as Opus 5.5, which achieved a higher 33.5% SecPass in 2.2 minutes for $116. Anti-cheating analysis found 29 problematic instances involving training recall or workspace leakage, and manual review reduced Argon’s initially claimed four unique security successes to two after rejecting cases affected by a dataset-description conflict and leaked test-environment artifacts. The results suggest that Argon’s deeper reasoning can enable difficult, independently derived security repairs, but its current efficiency limits its overall competitiveness, while the broader benchmark indicates that even leading AI coding agents still fail to secure most vulnerability-related tasks.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.