Google releases Gemini 4 Argon as independent benchmarks confirm strong results
Original title[AINews] AI Engineer NYC: Gemini 4 Argon, Vals AI Benchmarks
AISummary
Google made Gemini 4 Argon generally available for coding, knowledge work, enterprise workflows, and cyber defense.
Artificial Analysis places it level with GPT-6 Astra at roughly 60% of Astra's cost per task under introductory pricing, which doubles after November 30.
The digest also reports that Argon's real-world reliability is uneven, citing Theo's over-50% false-claim rate on a codebase-analysis benchmark.
AIWhy it matters
The digest pairs Gemini 4 Argon's launch with independent benchmarks and caveats on cost, memory, and reliability, which helps readers separate vendor claims from measured results.
Source: Latent Space · latent.spacePublished · added here