Gemini 4 Argon Tops Vals Index

Argon looks like Google’s strongest frontier-model comeback in years, but the leaderboard suggests a broadly capable model rather than an uncontested champion.

• –Ranks first on the Vals Index and Finance Agent v2, while tying for first on IOI with a perfect score.
• –Places second on Vibe Code Bench, Code Migration, and CyberBench, giving it unusually broad developer relevance.
• –Weak spots remain: 15th on MedScribe, seventh on CUA-bench, and fifth on Terminal-Bench 4.0.
• –Vals reports a $15.68 cost per composite test, below the leading Claude models, though long agentic tasks become substantially more expensive.
• –Broader developer validation will have to wait because Google is initially limiting access to trusted cyber defenders and selected testers.