TL;DR: Google globally released Gemini 4 Argon on Oct. 13, 2026 for developers, enterprises, and Pro and Ultra subscribers, claiming frontier results across coding, finance, legal, science, and cybersecurity.
Google ended months of waiting on Oct. 13, 2026, with the broad release of Gemini 4 Argon, its first major frontier model since Gemini 3 in late 2025. According to reporting by The Verge in “Google finally releases Gemini 4 Argon,” the company says the system delivers frontier reasoning across complex workflows while reducing hedging, repetition, and chatbot cliches.
That article, published Oct. 13 at 4:00 PM UTC, describes a model Google says can compete with the industry’s best and outperforms rival systems including GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 on benchmarks covering coding, cybersecurity, science, finance, and law. Google’s own announcement, “Introducing Gemini 4 Argon” from Sept. 30, 2026, and its follow-up “Gemini 4 Argon is now available for everyone” from Oct. 13, 2026, frame the same claim in enterprise terms: sustained reasoning for long-horizon work in software engineering, legal and financial research, and defensive cybersecurity.
Availability is broad but tiered. According to The Verge, Argon rolled out Oct. 13 to higher-tier Pro and Ultra subscribers in the Gemini app and AI Mode in Search, plus Ultra subscribers on the agentic development platform Antigravity. Google’s Oct. 13 announcement states the model is globally available for developers and enterprises through the API in AI Studio and the Gemini platform, for Ultra subscribers in Antigravity, and for Ultra and Pro subscribers in the Gemini app and AI Mode in Search. The Verge notes it remains unclear when — or whether — Argon will reach Plus subscribers or free users.
Why Google waited, then moved fast
The public launch followed a limited introduction. Google announced Argon on Sept. 30, 2026 as rolling out first to trusted cyber defenders through its Fairwind Program, according to both The Verge’s Sept. 30 coverage and Google’s “Introducing Gemini 4 Argon.”
Google said a phased method was needed to release frontier capability safely. In that September post, SVP of Google DeepMind and Chief AI Architect Koray Kavukcuoglu said the company was engaged in the U.S. government’s voluntary process for pre-release model access while it expanded access gradually. The company said it would collect feedback from early testers as it refined guardrails before wider availability for developers, enterprises, and consumers.
That caution reflected the model’s described cyber capability. Google said Argon was trained to find, validate, and patch critical software flaws on its own, and that trusted defenders and internal teams would receive a version without cyber guardrails to use full defensive power. The Verge’s Sept. 30 report summarized that approach as strengthening critical frontier safeguards before broader rollout, including defenses against misuse and prompt injection and monitoring for misalignment.
The timing mattered competitively. The Verge’s Oct. 13 story notes Argon arrived just a day after Meta announced its frontier model Muse Wonder Full, and after an interval in which systems from Anthropic and OpenAI left Google trailing. That same report describes seismic internal shakeups that raised questions about Google’s capacity to compete at the frontier. The launch, it concludes, shows the contest remains open and far from a two-company race.
What you pay and where you can use it
Pricing is positioned as aggressive. According to The Verge, Argon debuted at an introductory rate of $2 per million input tokens and $10 per million output tokens, described as roughly one-fifth the cost of OpenAI’s GPT-6 Astra and Anthropic’s Fable 5 models. Google’s October announcement gives the same $2 and $10 introductory figures, with cached input tokens at 95% off.
The discount is temporary. The Verge reports costs double at the end of November when the introductory period finishes, while developer documentation for “What’s new in Gemini 4 Argon” states introductory pricing runs through Nov. 30, 2026, after which list prices become $4 per million input tokens and $20 per million output tokens. The Verge cautions that per-token price does not guarantee lower bills in practice, because different systems may consume different token counts for equivalent work.
On the technical side, developer documentation identifies the system as gemini-4-argon with a default thinking level of medium and tunable options of low, medium, and high. It supports a unified 1M-token context window, with input, reasoning, and output sharing 1,048,576 tokens. Long decode allows a single response to extend well beyond the prior 64K-token ceiling, up to 1M tokens minus input length, using background execution for very long generations. Agentic video understanding is generally available, and the Antigravity agent in Managed Agents now supports Argon.
Benchmarks Google cites
Google centers its performance case on workplace and economic tasks rather than chat alone. Its September announcement says Argon reached 77.9% on DeepSWE v1.1, described as a measure of real-world long-horizon software engineering work. That same post calls the result a new state of the art for that test.
The company also highlights the Vals Index, which it describes as weighting finance, coding, legal, and tax work by contribution to U.S. GDP. According to Google’s October post, Argon ranks No. 1 on that index. The September post adds leading results on Vals Finance Agent v2 for multi-step financial research and Harvey’s Legal Agent Benchmark for legal research and drafting.
For business automation, Google cites 51.3% and No. 1 placement on AutomationBench, identified as Zapier’s test of end-to-end execution across core business functions. For visual understanding, Google cites 91.7% and state-of-the-art placement on LVBench for long video understanding, alongside claimed strength in professional chart analysis and action across document series. The Verge summarizes Google’s broader claim as outperformance across coding, cybersecurity, science, finance, and law against GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5.
Inside Google: quantum, memory, and code migration
Google says thousands of employees used Argon before public release for coding, deeper research, and writing quality. The September announcement details three internal research and infrastructure examples.
First, quantum algorithmic optimization. Google says Argon helped quantum researchers reduce spacetime resources — qubits multiplied by gates — for subroutines that limit important applications, beating a published baseline by 40% in minutes in one case.
Second, memory efficiency. A team of Argon agents examined fleet-wide profiling telemetry and applied memory optimizations across data centers, freeing more than 300 TiB of memory once deployed, with total projected savings of 500 TiB to 1 PiB.
Third, large-scale migration from C and C++ to Rust. Google says Argon agents are working on conversions ranging from tens of thousands of lines in core libraries such as re2 and libgav1 up to more than 800,000 lines for the Fuchsia Zircon kernel. The company stresses such rewrites face rigorous automated and manual auditing, emulation testing, and review before production. For libgav1, Google’s open-source video decoding software, agents replaced 32,000 lines of SIMD code after repeated profile-guided experiments, producing safe Rust that allowed automatic vectorization by the compiler. The result, Google says, is a memory-safe decoder running 2.7 times faster than the prior Rust port with identical video output, nearer the optimized C++ version.
Chrome, Python tooling, and open science speedups
The October announcement shifts from research pilots to shipped engineering work, reporting that Argon in Antigravity helped teams lift agentic code submissions by more than 35% while cutting code rollbacks by nearly one-third over recent weeks.
One case involved Chromium. Argon helped the Chrome team isolate core webpage handling from a legacy folder of about 6 million lines inside Chromium’s 40-million-line codebase. By handling multiple clean changes at once across hundreds of files, Google says the system compressed months of architectural refactoring into weeks. For Chrome’s billions of global users, the company says the change will let live pages shift among tabs, side panels, and floating windows without reloading and remove an entire class of browser crashes.
Another case involved Google’s open-source Agent Development Kit for Python, a framework for enterprise agents. The team supplied Argon with its agent skills rulebook and a one-paragraph objective to remove framework overhead without breaking applications. Working overnight on its own, the model proposed changes now live in google/adk-python that reduced in-memory session preparation time by 40% in long sessions and accelerated cold imports by 21%.
In open science, researchers used Argon in Antigravity to accelerate xTB, a widely cited open-source quantum chemistry package for predicting molecular 3D shapes and screening drug and materials candidates. Google reports gains of up to 2.3 times faster molecular structure prediction and 1.8 times faster conformer search, with changes contributed to the official xTB repository on GitHub.
The company also reports a 2.85-times end-to-end inference speedup for its open Gemma 4 E2B model on an A100 using vLLM, rising from about 37,900 to more than 108,000 tokens per second with full token parity and no quality loss in days rather than months.
YouTube, Wing, and defensive security pilots
Beyond code refactoring, The Verge reports Google is already applying Argon to infrastructure and robotics tasks, including halving the computing power required for YouTube’s comments section and debugging the drone physics simulator for Wing, Alphabet’s drone delivery company. The same report lists finding and patching cybersecurity flaws among internal uses.
Security partners received earlier access. Google’s September post says Wiz was already using Argon for defensive work through its Scan for Good initiative, described as a program to protect critical public infrastructure at no cost by locating and fixing high-risk exposures. In an early demonstration, Google says the model uncovered a critical flaw exposing sensitive personal information.
The October Google post broadens that picture to enterprise and research use: refactoring large codebases, improving legal, financial, and scientific workflows, patching vulnerabilities, and supporting advances in quantum computing, nuclear fusion, and life sciences.
Output scale built for marathon tasks
A central technical change is output capacity. Google says it expanded Argon’s output limit to 1M tokens, up from the previous 64K tokens. When a model can reason through hundreds of thousands of tokens in one trajectory, the company argues, it can maintain depth and solve harder problems without stopping and restarting.
That capacity connects to agentic development in Antigravity, where Google describes slash commands such as /goal and multi-agent loops with custom skills. Internal teams used those patterns to prototype 3D visualizations, games, and interactive animations in vanilla browser-renderable code. Examples cited include a detailed pagoda temple complex sculpted in Blender through visual critique loops, a playable miniature racing game built from desk photographs with Blender MCP, Nano Banana textures, and Three.js physics, and a procedurally generated 3D game with shaders, shatter physics, and single-player AI.
One demonstration pushed autonomy further: with no external libraries and pure Python, Argon wrote its own 3D ray-tracing engine with custom video and audio tools to construct and document a sunlit library, complete with narration and music generated by related Google systems.
What remains uncertain
Several questions remain open. The Verge says availability for Plus subscribers or free users is not clear. Pricing after November will be higher, and real-world cost will depend on token consumption, reasoning depth settings, and caching behavior. Competitive benchmark claims, while detailed, come from Google-selected tests and vendor descriptions; independent replication and developer experience over longer projects will test whether lab scores translate into durable productivity.
For now, the verified shift is access. After a September preview restricted to trusted defenders under the Fairwind Program and government pre-release review, Argon became globally available Oct. 13 through APIs, Antigravity, the Gemini app, and AI Mode in Search for qualifying subscribers. That move returns Google to the frontier conversation it had watched from the sidelines since late 2025.