Google opens Gemini 4 Argon to wide release and gives 10,000 researchers free access

Google today launched Gemini 4 Argon, its new flagship model, globally to developers and organizations through the API in AI Studio and the Gemini platform, to Google AI Ultra subscribers in Google Antigravity, and to Ultra and Pro subscribers in the Gemini app and AI Mode in Search. The company says Argon is built for deep work on complex, real-world tasks in domains such as code, finance, legal, cyber, and sciences, and it ranks first on the Vals Index, a benchmark that measures performance in critical economic sectors including finance, code, legal work, and tax.

The model debuts at an introductory price of $2 per million input tokens and $10 per million output tokens, with a 95 percent discount on cached input tokens. Google reports that Argon leads other frontier models on cross-domain benchmarks, though it has not yet published full performance metrics against specific competitors such as GPT-4o or Claude 3.5 Sonnet, only the Vals Index ranking.

Google teams are already running Argon in production on their hardest problems. In Chrome, the model helped separate the core that handles web pages from a legacy folder of roughly 6 million lines of code inside the 40-million-line Chromium codebase — an architectural refactoring spanning hundreds of files that finished in weeks instead of months. The result: live web pages can now move between tabs, side panels, and floating windows without a reload, eliminating an entire class of browser crashes.

In the Agent Development Kit (ADK), an open-source framework that powers enterprise agents, Argon was given the agents/skills folder rulebook and a single-paragraph goal: cut overhead without breaking existing applications. Overnight autonomous work produced optimizations now live in google/adk-python: a 40 percent reduction in in-memory session preparation time for long sessions, and a 21 percent acceleration of cold imports. In the xTB quantum chemistry package, one of the most-cited tools for predicting 3D molecular structures and screening drug and material candidates, Argon sped up structure prediction by 2.3x and conformer search by 1.8x; the changes were contributed back to the official GitHub repository.

On A100 GPUs via vLLM, Argon achieved a 2.85x end-to-end speedup in Gemma 4 E2B inference, lifting throughput from roughly 37,900 to over 108,000 tokens per second with 100 percent token parity and no quality loss, in a process that took days rather than months. The optimizations, ranging from attention tuning to speculative decoding, were submitted as pull requests upstream.

To accelerate research with Argon, Google is offering 10,000 scientists free access to the highest usage tier in Google Antigravity for six months. Sign-up details are available in the link attached to the official announcement.