Google DeepMind Launches Gemini 4 Argon Frontier Model

Google DeepMind released Gemini 4 Argon globally on Oct. 13, 2026, making its new flagship model available to developers and enterprises through the Gemini API in AI Studio, the Antigravity platform, and the Gemini app and AI Mode in Search. The launch drops in the middle of one of the most compressed frontier-model release cycles in recent memory, following Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6.1 Sol by just a matter of weeks.

Argon is built for deep, multi-step professional work — the kind of task that breaks less capable models: large-scale codebase refactoring, longitudinal financial analysis, legal document synthesis, cybersecurity vulnerability discovery and scientific computation. According to Google DeepMind, it ranks first on the Vals Index, an industry benchmark that evaluates model performance across finance, coding, legal and tax work, scoring 68.90% versus Claude Sonnet 5.5 at 67.04% and Claude Opus 5.5 at 66.97%.

“Argon excels at deep work in complex, real-world tasks across domains including coding, finance, legal, cybersecurity, and science,” wrote Koray Kavukcuoglu, SVP at Google DeepMind and chief AI architect at Google.

The proof points Google offers are unusually concrete. Internal teams using Argon through Antigravity saw agentic code submissions increase by more than 35% while cutting code rollbacks by nearly a third. The Chrome team used it to complete a major architectural refactor across hundreds of files in Chromium’s 40-million-line codebase — work that would have taken months — in a matter of weeks. On the science side, Google used Argon to make the widely-used xTB quantum chemistry package up to 2.3 times faster at predicting molecular structures, and to accelerate inference on its open Gemma 4 models by 2.85 times on an A100 GPU, boosting throughput from roughly 37,900 to more than 108,000 tokens per second.

How Argon Stacks Up Against Claude and GPT-6

The competitive picture is genuinely nuanced. Argon leads on the Vals Index and beats Claude Opus 5.5 on the DeepSWE v1.1 software engineering benchmark (77.9% vs. 74.2%). But it trails Opus 5.5 on Terminal-Bench 4.0 (57% vs. 60%), which measures autonomous terminal-based coding — the kind of interactive CLI work where Anthropic’s models have built their strongest reputation among working developers. In short, Argon appears to be the better model for structured, domain-specific professional tasks; Claude remains more competitive for free-form, iterative coding sessions.

Where Argon is unambiguously ahead is on price. The introductory API rate is $2 per million input tokens and $10 per million output tokens — with cached input tokens at 95% off. That compares favorably to GPT-6 Astra and Claude Fable 5.1, both priced at $10 per million input tokens and $50 per million output tokens. Google has flagged that introductory pricing is set to double on Dec. 1, 2026, which creates a real incentive to build and test now. A standout technical differentiator is Argon’s 1-million-token maximum output limit, which community observers have noted could matter significantly for long-running autonomous agents that need to generate large artifacts in a single pass.

What This Means for Students and Early-Career Professionals

The honest answer on access is that Argon is paywalled for most students. In the Gemini app and AI Mode in Search, only Google AI Pro ($20/month) and AI Ultra (starting at $99/month) subscribers can select the model; free accounts are excluded. That said, there are two meaningful exceptions worth knowing about. First, Google is offering 10,000 selected scientists six months of Antigravity’s highest usage tier at no cost — a program that graduate students and STEM researchers should actively look into. Second, Argon is available inside GitHub Copilot (Pro+, Max, Business and Enterprise plans) and is listed in Cursor’s model documentation, which may already be covered for students through university licenses or existing subscriptions to coding tools.

For those building on the API directly, the window before the Dec. 1 price increase is a practical reason to experiment now. Argon’s demonstrated strengths map well onto the kinds of work students use to build portfolios: synthesizing large technical documents, navigating complex codebases, and producing domain-specific analysis in law, finance or science. Whether you’re working on a capstone project, thesis research, or a side app to show future employers, those use cases align with what Argon is demonstrably good at — not just what Google’s marketing says it can do.

The bottom line: Argon is a serious model that earns its flagship label on several key benchmarks, launches at a price point well below comparable rivals, and has a brief window of introductory pricing that makes API experimentation unusually affordable. Students and developers already paying for Cursor or GitHub Copilot can access it today without any additional cost.

Source: Google DeepMind