Google Has Introduced Gemini 4 Argon, and Here Is Everything You Need to Know
Gemini 4 Argon targets cybersecurity, coding, enterprise work with $2/$10 per million input/output, 1 million token limit, and top benchmarks.
Key Takeaways
- Google has released Gemini 4 Argon, a new frontier model focused on deep reasoning for long, complex workflows, initially offering it to select cybersecurity defenders through the Fairwind Program at $2 per million input tokens and $10 per million output tokens.
- The model is already being used internally at Google by thousands of employees for coding, research, and writing, including quantum computing optimization, data center memory savings, and large-scale C/C++ to Rust migration.
- Argon’s output limit has been expanded to 1 million tokens, enabling it to handle more complex tasks in a single trajectory rather than requiring many smaller steps.
- Google reports leading benchmark results across software engineering, enterprise knowledge work, finance, legal tasks, automation, and long-video understanding.
- The model is particularly strong in defensive cybersecurity, where it can autonomously find, validate, and patch vulnerabilities, and is being released without cyber guardrails for trusted defenders while broader safeguards are prepared for public access.
Google has released Gemini 4 Argon, its newest frontier model, to a select group of cybersecurity defenders through the Fairwind Program, with an introductory price of $2 per million input tokens and $10 per million output tokens. Cached input tokens are priced at 95% off the input token rate. The model is designed to handle long, complex workflows that require sustained reasoning, and Google says it is already being used internally across thousands of employees for coding, research, and writing tasks.
What Argon Is and How It Is Being Released
Gemini 4 Argon is built for deep reasoning across extended, multi-step tasks. Google describes it as delivering frontier performance in real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense. The company says it engaged in the U.S. government's voluntary pre-release model access process while gradually widening access. Feedback from early testers will inform further work on guardrails before Argon reaches developers, enterprises, and consumers.
The release follows a phased pattern: trusted cyber defenders first, then broader availability. Google has not given a date for general access.
Internal Use at Google
Argon is already powering workflows inside Google. Thousands of employees have used it for specialized coding, deeper research, and writing quality, according to the company. Three examples stand out.
In quantum computing, Argon helped researchers optimize spacetime resources, measured as qubits multiplied by gates, for subroutines that bottleneck important applications. In one case it beat the published baseline by 40% in a matter of minutes.
A team of Argon agents analyzed fleet-wide profiling telemetry and autonomously identified and applied memory optimizations across Google data centers. Once rolled out, this freed over 300 TiB of memory, with estimated total savings of 500 TiB to 1 PiB.
Argon agents are also working on migrating C/C++ codebases to Rust across Google, scaling 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. Because many of these systems are critical, the rewrites undergo automated and manual auditing, emulation testing, and review before production rollout.
For libgav1, Google's open source video decoding software, Argon agents took an existing Rust port and replaced 32,000 lines of SIMD code by running many rounds of profile-guided experiments, studying compiler output, and producing safe Rust so the compiler would vectorize it automatically. The result is a memory-safe video decoder that runs 2.7 times faster than the Rust port with identical video output, bringing it closer to the optimized C++ version.
A Much Larger Output Limit
To support longer and more complex use cases, Google has expanded Argon's output token limit to an industry-leading 1 million tokens, up from the previous 64,000. The company argues that when a model has room to think deeply and generate hundreds of thousands of tokens in a single trajectory, it can solve tough problems in one go rather than across many smaller steps.
Benchmark Results Across Coding and Enterprise Work
Google reports that Argon sets a new state of the art on DeepSWE v1.1 with 77.9%, a benchmark measuring performance on real-world long-horizon software engineering tasks. It also leads on the Vals Index, which measures economic impact across finance, coding, legal, and tax work, with each sector weighted by its contribution to U.S. GDP. Google cites similarly leading results on Vals Finance Agent v2, which covers multi-step financial research, and Harvey's Legal Agent Benchmark, which covers legal research and drafting. On AutomationBench, Zapier's benchmark for end-to-end execution across core business functions, Argon ranks first with 51.3%.
The model is also strong where knowledge work requires visual understanding. It can drive professional chart analysis, identify details from long videos, and act on a series of documents. On LVBench, which measures long video understanding, Argon is state of the art at 91.7%.
Cybersecurity Defense
Google trained Argon to be highly capable at defensive cybersecurity. The model can autonomously find, validate, and patch critical software vulnerabilities. For trusted defenders and Google's own internal teams, the company is releasing Argon without cyber guardrails so they can use its full frontier-level defense capabilities.
Wiz is already using Argon through its Scan for Good initiative, a program that protects critical public infrastructure for free by finding and remediating high-risk exposures. In an early demonstration, the model uncovered a critical vulnerability exposing sensitive personal information in healthcare software used by hospitals worldwide, a severe risk that previous frontier models had missed.
On CWE-bench v1, which evaluates vulnerability remediation, Argon ties for first place with a top score of 68%, building on 3.8 Flash Cyber's performance on CWE-bench v0. Google also reports gains in vulnerability discovery over 3.8 Flash Cyber: on its internal comprehensive vulnerability benchmark, Argon uncovered exposures across complex codebases spanning 20 programming languages, and on Wiz's internal black-box penetration testing benchmark, which tests analysis of live web systems without source code, Argon outperformed 3.8 Flash Cyber in discovering attack surfaces, identifying vulnerabilities, and producing proof-of-concept evidence.
Safeguards Before Wider Release
Google says it is strengthening safeguards in four areas before broadening access. Against misuse, Argon is designed to refuse harmful requests related to cyber or chemical, biological, radiological, and nuclear attacks while preserving legitimate dual-use scientific research, in line with Google's Frontier Safety Framework. The company says it improved techniques to monitor the model's internal activations to spot misuse, and that internal and external red teams tested these safeguards using manual and automated attack methods.
Against prompt injection, where malicious instructions hijack a model's behavior, Google calls Argon its most resilient model yet and says it leads on Gray Swan's Indirect Prompt Injection benchmark through automated red teaming and adversarial training.
For misalignment, Google is deploying mitigations that monitor Argon's chain-of-thought and actions and stop execution when necessary. A similar system monitored training runs and alerted a dedicated incident response team, with precautions taken not to feed findings back into training in ways that could shape Argon's reasoning to evade monitoring. Google encourages the rest of the industry to preserve reasoning transparency as capabilities increase.
For system hardening, Google says it is isolating and sealing sandboxed environments before high-risk training or evaluations begin, in line with its agent control roadmap, and will share agent security best practices with partners.
Professional and Scientific Work
Google says Argon combines frontier reasoning, deep visual perception, expanded domain knowledge, and stronger writing across legal, finance, science, and coding. It answers more directly with less hedging, avoids unnecessary repetition, skips clichés, and follows style instructions better than previous Gemini models, according to the company.
In quantum computing, Argon translated an internal version of AlphaQubit, Google's machine learning quantum error correction decoder, into a single highly optimized hardware kernel, nearly doubling inference throughput so larger models can be deployed and scaled to larger quantum devices while meeting strict real-time thresholds. For arcsin, a building block in recent quantum simulation algorithms, an implementation written by Argon and formally verified in Lean cut spacetime cost by more than 100 times compared with a published quantum CORDIC baseline.
In nuclear fusion, a Google team alongside Professor Matt Landreman of the University of Maryland discovered a broader class of counterexamples to Grad's conjecture, a long-standing hypothesis from 1967 that exact mathematical solutions for stellarator magnetic fields were impossible. The new solutions give the fusion community reliable benchmarks for testing stellarator simulation codes.
In rare disease research, Gamow Labs used Argon to analyze de-identified genomes from newborns with rare developmental lung disorders. In one case, Argon pinpointed a chromosomal rearrangement that routine software had missed for years and explained why it went undetected. On Gamow Labs evaluations derived from real clinical and lab data, Argon scored 57% on RareBench for rare-disease diagnosis assistance and 52.2% on LabBench for wet-lab reasoning.
What to Watch
The key questions are when Argon becomes generally available and how its pricing holds after the introductory period. Google has framed the release as gradual, with guardrails still being iterated, so broader access will likely depend on how the safeguards perform under outside testing. For enterprises, the 1 million token output limit and the benchmark results in coding, finance, and legal work are the numbers worth tracking, while the cybersecurity results will draw attention from defenders who have wanted models without cyber guardrails for authorized work.

Comments (0)