Gemini 4 Argon is now globally available to developers and enterprises via the Gemini API and AI Studio, to Google AI Ultra subscribers in Google Antigravity, and to Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search.
Argon is designed for deep work on complex, real-world tasks across domains such as coding, finance, legal, cybersecurity and science. Google says Argon achieves superior depth and precision on multi-step tasks, ranking #1 on the Vals Index. Introductory pricing is $2 per 1M input tokens and $10 per 1M output tokens, with cached input tokens discounted by 95%.
Early use cases and measured impacts
Teams at Google, enterprise customers and researchers report Argon is already driving measurable gains:
- Refactoring and optimizing large codebases
- Improving business workflows across legal, finance and science
- Finding and patching critical cybersecurity vulnerabilities
- Enabling advances in quantum computing, nuclear fusion and life sciences
Selected examples and metrics provided by Google:
- Software engineering: In Antigravity, Argon reportedly increased agentic code submissions by more than 35% and reduced code rollbacks by nearly one third.
- Chrome: Argon helped separate core webpage handling from an approximately 6 million-line legacy folder within Chromium’s 40 million-line codebase and completed an architectural refactor spanning hundreds of files in weeks instead of months.
- ADK (Agent Development Kit): Autonomous runs of Argon produced optimizations now in google/adk-python that reduced in-memory session prep by 40% in long sessions and sped cold imports by 21%.
- xTB quantum chemistry package: Argon-assisted improvements made xTB up to 2.3× faster at predicting molecular structures and 1.8× faster at conformer search; changes were contributed to the official xTB GitHub repository.
- Gemma models: Using vLLM, Argon-enabled optimizations achieved a 2.85× end-to-end inference speedup on an A100 (roughly from ~37,900 to over 108,000 tokens/sec on Gemma 4 E2B) with token parity and no quality loss.
- 3D visualizations and mini-games: Argon produced photorealistic 3D temple complexes and turned desk photos into playable miniature racing games by orchestrating Blender MCP, Nano Banana texturing and Gemini Omni Flash assets, then rendering to the web with Three.js.
Transforming professional workflows
Google positions Argon as stronger at frontier reasoning, multimodal document comprehension and clearer writing for high-stakes professional workflows. Reported uses include:
- Synthesizing ~50 GB of airborne LiDAR, GIS vectors, Library of Congress CAD drawings and archives to reconstruct 3D models of Chaco Canyon.
- In the Gemini App, formatting unstructured project folders into client proposal decks and automating future request workflows.
- At Isomorphic Labs, accelerating drug discovery by rapidly extracting relevant information from complex PDFs and patents to shorten processes that previously took weeks down to hours.
Scientific research and domain advances
Argon is reported to accelerate research across multiple scientific areas:
- Quantum computing: Argon translated an internal AlphaQubit ML decoder into a highly optimized hardware kernel, nearly doubling inference throughput, and produced a Lean-verified implementation for a quantum subroutine that cuts spacetime costs by over 100× versus a published CORDIC baseline.
- Nuclear fusion: Google and Professor Matt Landreman (University of Maryland) found a broader class of counterexamples to Grad’s conjecture, supplying benchmarks useful for stellarator simulation validation.
- Rare disease research: Gamow Labs used Argon on de-identified newborn genomes; on private evals Argon topped RareBench (57% on rare-disease diagnosis assistance) and LabBench (52.2% on wet-lab reasoning), including finding a chromosomal rearrangement missed by routine software.
- Enzyme design: Argon assembled data, tuned models and proposed five novel serine esterase designs for the 4MU-acetate benchmark; laboratory tests confirmed catalytic activity.
Google also announced it will offer 10,000 scientists free access to the highest usage tier of Google Antigravity for six months (onboarded on a rolling basis), subject to eligibility and capacity; applications are open.
Robotics and Video-to-Sim-to-Real
Combining visual-semantic reasoning, physical understanding and autonomous code generation, Argon targets robotics development. On the Embodied Reasoning (ERQA) benchmark, Argon scored 87%, outperforming other frontier models according to Google.
In an internal test, Argon watched raw video of a human playing cornhole, inferred the environment’s spatial coordinates and whole-body joint movements, generated simulation code, and transferred the result to an Apptronik Apollo humanoid that then sank seven consecutive shots in a real-world trial. Google describes this pipeline as compressing weeks of manual simulation work into hours.
Cybersecurity and Fairwind Program
Google is expanding Argon access to additional Fairwind Program partners. Wiz is using Argon in its “Scan for Good” initiative to protect critical infrastructure. Examples cited by Google include detecting and remediating a vulnerability that could have exposed sensitive government data and fixing a public file download bypass that exposed a master key and administrator access, protecting hundreds of thousands of private patient records.
Safety, misuse protections and monitoring
Google says Argon incorporates an extensive safety and security stack:
- Misuse and jailbreak defenses: Under the Frontier Safety Framework and external red teaming, Argon’s robustness improved — external bio red teaming unlocked less than 1% of tested bio sub-tasks. On adversarial tests for dangerous requests, violations were reduced nearly twofold versus prior Gemini models.
- Agent alignment and real-time monitoring: Argon was trained to respect system boundaries, report negative findings, and pause for human confirmation before irreversible actions. Shortcut attempts on coding benchmarks dropped to 2% on Gemini 4 Argon. An automated misalignment monitor reads step-by-step reasoning and actions to block unintended workarounds in real time.
- Prompt-injection resistance: Through adversarial training and automated red teaming, Argon leads the Gray Swan Indirect Prompt Injection (IPI) benchmark, reducing attack success across 15 adaptive attempts to 0.7%.
More details are available in the Gemini 4 Argon Model Card.
Availability and pricing
Argon launches today with a 1M unified input-output token limit via the Gemini API. Google published documentation for developers and enterprises on the continuation protocol and recommended setups.
Rollout and access:
- Developers: build with the Gemini API paid tier; Google AI Ultra subscribers can try Argon in Google Antigravity, AI Studio and Android Studio.
- Enterprises: access via Gemini for business, the API and Google Antigravity.
- Everyday users: Argon is accessible in the Gemini app and AI Mode in Search for Google AI Ultra or Pro subscribers.
- Scientists & researchers: applications are open for free six-month access to Antigravity’s highest usage tier for selected applicants.
- Cybersecurity defenders: apply to the Fairwind Program for access without cyber guardrails on a rolling basis.
Pricing note: after the introductory period ends on November 30, 2026, public-access pricing will change to $4 per 1M input tokens and $20 per 1M output tokens for the Argon model.
Conclusion
Gemini 4 Argon is presented by Google as a multimodal, high-context model aimed at enterprise-grade, scientific and engineering problems. Google highlights benchmark leadership, concrete speed and productivity gains across internal and partner projects, and a comprehensive safety and monitoring approach intended to reduce misuse while enabling high-impact research and operations.
