Google finally releases Gemini 4 Argon

Summary

Google has announced Gemini 4 Argon, its first frontier model since Gemini 3 at the end of 2025. The company claims the model outperforms GPT-6 Astra and Claude models in benchmarks. It also offers pricing approximately one-fifth of competitors' per-token costs. The model is being rolled out to Pro and Ultra subscribers via the Gemini app, Search AI mode, and the Antigravity platform. No specific timeline has been announced for Plus or free users.

Why it matters

After lagging behind OpenAI and Anthropic for months during internal organizational changes, Google proves it remains a reliable frontier player with Argon. Aggressive initial pricing ($2/$10 per million tokens) could pressure market cost structures and accelerate enterprise adoption of frontier models. The model's release just one day after Meta's Muse Wonder Full highlights that the frontier race extends far beyond a two-horse competition.

Highlights

• Starting price is approximately 80% lower than GPT-6 Astra and Fable 5; expected to double by the end of November
• Internal deployments include halving the computational load for YouTube comments and automatically patching security vulnerabilities
• Rollout is limited to Pro/Ultra tiers; no confirmed date for Plus and free users

Critical take

The article is essentially a near-verbatim repetition of Google's own press claims. Every performance assertion begins with 'Google says.' No independent benchmarks, third-party evaluations, or real user feedback are cited. A more rigorous critique should question whether the term 'frontier' is a marketing label or a defensible technical distinction. This inquiry becomes even more critical given the model's limited accessibility and the lack of explanation regarding the methodology behind the benchmarks.