Google and Meta ship frontier models the same week, closing the gap with Anthropic and OpenAI

Last week Anthropic and OpenAI still looked untouchable at the frontier. Google had only made incremental gains since Gemini 3 and weathered leadership churn: Jeff Dean leaving to found a startup, Demis Hassabis stepping back from day-to-day management, and the AI team moving from DeepMind to Google proper. On Tuesday Google released Gemini 4 Argon, claiming it matches or beats OpenAI's and Anthropic's flagships on multiple benchmarks, priced up to 80% cheaper per token during the promo period and still notably cheaper after. On Monday Meta released Muse Wonder Full, which by its own testing competes with GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1 — a step up from April's Muse Spark, which came out of Superintelligence Labs as "good but not frontier." For teams doing model selection and cost modeling, this is the first time comparable capability comes with a several-fold price gap, so API migration and eval costs need recalculating. For teams distributing through Meta's ecosystem, Muse's rapid user growth shows that a multi-billion-user channel is itself a moat. And for open-weights efforts, the piece notes open models keep getting stronger, making long-term leadership hard for anyone and shortening the frontier window.