Google Launches Gemini 4 Argon, Its New Frontier Model for Deep Work

Google DeepMind has made Gemini 4 Argon generally available, expanding access beyond the limited Fairwind Program for cybersecurity partners. Announced on September 30 and rolled out globally on October 13, 2026, the model is positioned as Google's latest frontier system for complex, multi-step tasks in coding, finance, legal work, cybersecurity, and science.

What Is Gemini 4 Argon?

Argon is the first model in the Gemini 4 generation. Google describes it as excelling at "deep work"—long-horizon reasoning and agentic tasks rather than short chat responses. It features a unified 1 million token context window shared across input, thinking, and output, along with support for long decode up to 1 million output tokens. Tunable thinking levels (low, medium, high) are available.

The model is accessible today through the paid Gemini API in AI Studio and the Gemini platform (model ID: gemini-4-argon), for Google AI Ultra subscribers in Google Antigravity, and for Google AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search. Free users and lower tiers continue to use Flash-class models.

Pricing and Availability

Launch pricing is introductory through November 30, 2026:

• Input tokens: $2 per million
• Output tokens: $10 per million
• Cached input tokens: 95% discount ($0.10 per million)

After November 30, rates double to $4/$20. Google notes this is competitive relative to peers such as OpenAI’s GPT-6 Astra and Anthropic’s Claude models. Early enterprise testers include Deloitte, LexisNexis, Ambience Healthcare, and Isomorphic Labs.

Sundar Pichai confirmed the global rollout on X, pointing to the official blog for full details.

Performance Claims

Google reports Argon ranks first on the Vals Index, which weights performance across finance, coding, legal, and tax work by economic contribution. It also leads on DeepSWE v1.1 (77.9%) for long-horizon software engineering, Zapier’s AutomationBench (51.3%), LVBench for long video understanding (91.7%), and ties for first on CWE-bench v1 (68%) for vulnerability remediation.

Independent trackers such as Artificial Analysis place it near the top of the Intelligence Index, competitive with GPT-6 Astra. Google says internal use has increased agentic code submissions by more than 35% while reducing rollbacks by nearly a third in Antigravity.

Implications for Developers and Businesses

For developers, Argon’s long context and strong coding performance make it attractive for complex software engineering agents and multi-step workflows. The aggressive introductory pricing lowers the barrier for experimentation compared with higher-cost frontier models.

Enterprises gain a tool positioned for knowledge work that requires precision—legal research with citation fidelity, financial analysis, and cybersecurity defense. The phased release (cyber partners first, then broader access) reflects ongoing safety considerations around frontier capabilities.

The model’s visual understanding strengths also open applications in document analysis, chart interpretation, and video-based tasks.

Why It Matters

Gemini 4 Argon’s general availability marks Google’s return to the leading edge of the frontier model race after a period of more limited releases. By combining strong benchmark results, a 1M-token window, and lower introductory pricing, it gives developers and companies a practical new option for agentic and deep-work applications. How it performs in real production settings versus Claude and GPT-class systems will be closely watched in the coming weeks.

---

Sources: Google Blog – Introducing Gemini 4 Argon, Google Blog – Gemini 4 Argon availability, Reuters, The Verge