Google has officially launched Gemini 4 Argon, its new flagship AI model, making it globally available to developers, businesses, and consumers. This update redefines access tiers and brings advanced AI capabilities, including a focus on coding and 3D generation.
Google has officially rolled out Gemini 4 Argon, its new flagship AI model, making it accessible to developers, businesses, and consumers worldwide starting today. This is a significant move that establishes a new top tier in Google's AI offerings.
Users can access Gemini 4 Argon through the API in AI Studio and the Gemini platform, according to Google. If you're an AI Ultra subscriber, you'll find it in Google Antigravity. Both AI Ultra and Pro subscribers can also use the model via the Gemini app and Google Search's AI mode.
New Tiers and Pricing Details
This launch brings a significant shake-up to Google's AI model tiers. Free Gemini users and existing paying Plus subscribers will now only have access to the weaker Flash models. Gemini 4 Argon takes over from the Gemini Pro class as Google's new 'frontier series,' positioning it as the top-tier offering.
Google has stated that more Gemini 4 models are on the way, likely meaning updated Flash variants that will eventually reach free and Plus users. Internally, Google is reportedly already testing a potential successor codenamed "Carbon," which is supposed to outperform Argon at coding in particular.
For now, the pricing for Argon stands at $2 per million input tokens and $10 per million output tokens at an introductory rate. There's also a 95 percent discount for cached inputs, which is a good bit of savings for repeated queries. While this pricing is in US dollars, it gives Indian developers and businesses a clear idea of the cost involved. Direct conversion might make it seem steep for some, but for serious AI applications, it's the new standard from Google.
Performance and What It Means for Users
On the performance front, Gemini 4 Argon currently holds the top spot on the Vals Index. However, it still trails behind Anthropic's Claude Opus 5.5 in independent overall rankings, as reported by The Decoder. Google's brief delay in shipping also means it's now behind Meta's Muse Wonder Full, which launched just yesterday.
Google shared some internal figures, though these are, of course, self-reported. The company claims software engineers using Argon saw over a 35 percent increase in agent-driven code submissions, with code rollbacks dropping by nearly a third.
The real-world applications highlighted by Google are quite impressive. In the Chromium project, Argon helped the Chrome team complete a major architectural refactor across hundreds of files in their massive 40-million-line codebase. Work that would have typically taken months was finished in just a few weeks. For web users, this means web pages can move between tabs, sidebars, and floating windows without needing to reload, which is a significant improvement.
Working with Google's open-source Agent Development Kit, Argon also autonomously found optimizations overnight. It cut session preparation time by 40 percent and sped up cold imports by 21 percent. The inference speed for open Gemma models on an A100 via vLLM saw a 2.85-fold jump, going from approximately 37,900 to over 108,000 tokens per second.
Advanced Capabilities and Safety
Beyond code, Argon can generate complex 3D models, games, and animations purely in code that run directly in the browser. One demo showcased a raytracing engine written entirely in Python without external libraries. Google even published the source code for this, which is great for transparency and learning.
In another demonstration, Argon started from a single image in Antigravity and controlled Blender autonomously. It modeled gothic architecture and set up audio-reactive lighting to produce a finished 3D scene, Google says. This capability is notable, especially since Meta AI chief Alexandr Wang has indicated that Meta's Muse models were not specifically trained for 3D demo games.
Google is also touting Argon's architecture as its strongest to date in terms of safety. External red teaming in the bio domain bypassed fewer than 1 percent of tested safety checks, and violations on an adversarial set dropped by nearly half compared to earlier Gemini models. For agent alignment, attempts to take unauthorized shortcuts fell to just 2 percent, thanks to an automated misalignment monitor.
Argon achieved the lowest attack rate on the Gray Swan Indirect Prompt Injection benchmark, with a successful attack rate of only 0.7 percent across 15 adaptive attempts. For comparison, GPT-6 Astra hit 8.5 percent and Claude Opus 5 came in at 4.8 percent in a previous Gray Swan test, though Google's write-up doesn't confirm if the exact same tests were used.
Access for Researchers and Cybersecurity
On the access front, Google is offering 10,000 researchers the highest usage tier of Google Antigravity free for six months. Applications are open now, with spots awarded on a rolling basis depending on eligibility, capacity, and selection. Cybersecurity defenders can also apply separately for the Fairwind program, which grants Argon access without cyber safeguards. Google plans to expand this program to hundreds of additional partners.
Google's Gemini 4 Argon marks a significant step for the company in the competitive AI space. While its pricing and current benchmark position against rivals like Claude Opus 5.5 and Meta's Muse Wonder Full are key considerations, the demonstrated capabilities in coding, 3D generation, and safety improvements are compelling. For Indian developers and businesses looking to integrate advanced AI, Argon presents a powerful, albeit premium, new option.
