Google Releases Gemini 4 Argon With 1M Output Limit and Cyber Focus
Google claims the flagship model introduces an expanded output window, autonomous security auditing, and deep reasoning across enterprise and coding workflows.
Key takeaways
- Gemini 4 Argon expands the model output token limit to 1 million tokens, up from the prior 64,000 token maximum.
- The model achieved a 77.9% score on the DeepSWE v1.1 coding benchmark and tied for first place on CWE-bench v1 with a 68% score.
- Google is conducting a phased rollout via its Fairwind Program and voluntary U.S. government review before widening access to API and Google AI Ultra subscribers.
- Introductory API pricing is set at $2 per million input tokens and $10 per million output tokens, before shifting to standard rates of $4 and $20.
On September 30, 2026, Google officially launched its flagship frontier model, Gemini 4 Argon, targeting complex multi-step reasoning across software engineering, enterprise operations, and cybersecurity defense. The release expands output capacity and autonomous task capabilities. The company revealed full technical benchmarks, enterprise use cases, internal deployment results, and pricing tiers as reported on the official Google Blog.
Gemini 4 Argon focuses on sustained trajectories for long-horizon workflows. To support these prolonged analytical runs, Google expanded the maximum output window to 1 million tokens, a substantial increase from the prior 64,000-token ceiling. According to Google, having the headroom to generate hundreds of thousands of tokens in a single trajectory allows the model to add a new level of depth in reasoning to solve tough problems in one go.
Technical Performance and Benchmark Results
Google designed Gemini 4 Argon to handle complex workflows that require deep reasoning and multimodal understanding. According to evaluation data published by 9to5Google, Gemini 4 Argon set a new state-of-the-art mark on the DeepSWE v1.1 benchmark for real-world software engineering with a score of 77.9%. This placed it ahead of Claude Opus 5.5 (74.2%) and OpenAI's GPT-6 Astra (74.1%).
On broader enterprise metrics, Google reported top placements on the Vals Index, which evaluates economic task execution across legal, finance, tax, and coding weighted by U.S. GDP contribution. According to Google's announcement, the model also ranked first on Zapier's AutomationBench with a score of 51.3% for end-to-end execution of core business operations. For multimodal visual understanding, Google highlighted separate chart analysis and document processing capabilities, while noting a 91.7% score on LVBench specifically for long video understanding.
Independent testing from benchmarking firm Artificial Analysis, reported by Engadget, found that Gemini 4 Argon matched GPT-6 Astra on its Intelligence Index at approximately 60% of the cost per task under introductory pricing. Artificial Analysis also measured a 15% hallucination rate for Argon, compared to 54% recorded for both GPT-6 Astra and GPT-6.1 Sol.
Internal Google Workflows and Codebase Migrations
Google demonstrated Argon's real-world utility by integrating it directly into internal engineering and infrastructure operations. Google reported that thousands of its engineers have used Argon internally for specialized coding tasks, conducting deeper research, and writing quality.
In quantum computing research, Google reported that Argon helped optimize spacetime resources (qubits multiplied by gates) in computational subroutines, beating the published baseline by 40% in minutes in one example. In infrastructure management, Google stated that autonomous Argon agents analyzed fleet-wide profiling telemetry across its data centers, applying memory optimizations to free up over 300 TiB of memory once rolled out, with an estimated 500 TiB to 1 PiB in total savings.
Google also reported that Argon agents are working on migrating C and C++ codebases to Rust across Google, scaling from core software libraries like re2 and libgav1 up to 800,000+ lines for the Fuchsia operating system's Zircon kernel. According to Google, for its open-source video decoder libgav1, Argon agents replaced 32,000 lines of SIMD code with safe Rust through iterative profiling experiments, producing a memory-safe decoder that runs 2.7 times faster than the Rust port while maintaining identical video output.
Cybersecurity Capabilities and Guardrails
Defensive cybersecurity represents a central focus for Gemini 4 Argon. On the CWE-bench v1 vulnerability remediation benchmark, Argon tied for first place alongside Grok 4.7 and GPT-6 Astra with a score of 68%, as reported by Engadget.
Google stated that cybersecurity partner Wiz deployed Argon through its Scan for Good initiative to protect critical public infrastructure. During testing, the model discovered a severe vulnerability in healthcare management software used internationally by hospitals that had gone undetected by previous models, as documented by The Guardian. According to internal Google benchmarks across 20 programming languages and Wiz black-box penetration tests, Argon demonstrated significant improvements in vulnerability discovery and proof-of-concept validation over 3.8 Flash Cyber.
To manage potential security risks, Google is rolling out Argon without cybersecurity guardrails exclusively to vetted defenders and internal teams, while enforcing strict restrictions for broader access. As reported by The Guardian, Google chief AI architect Koray Kavukcuoglu noted that releasing frontier capabilities safely requires a phased approach. The announcement coincided with voluntary safety commitments made during White House discussions with technology executives.
Safety mitigations outlined by Google include:
- Misuse Defenses: Filtering mechanisms and internal activation monitoring to prevent exploitation for cyber, chemical, biological, radiological, or nuclear (CBRN) harm while permitting legitimate dual-use scientific research.
- Prompt Injection Resilience: Defense layers tested on the Gray Swan Indirect Prompt Injection (IPI) benchmark to counter malicious context hijacking.
- Misalignment Monitoring: Chain-of-thought tracking to stop unauthorized autonomous action runs without corrupting the core training data.
- System Hardening: Isolated, sealed sandboxes for high-risk training and evaluation runs.
Pricing and Phased Availability
Gemini 4 Argon is initially restricted to members of Google's Fairwind Program, which serves governments and specialized cybersecurity partners, alongside pre-release U.S. government safety reviews. Broader access will roll out progressively to paid API customers and Google AI Ultra subscribers.
According to 9to5Google, Google established an introductory API price of $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted by 95%. Once the introductory window concludes, standard API pricing will adjust to $4 per million input tokens and $20 per million output tokens.
Frequently asked questions
What is the output token limit for Gemini 4 Argon?
Gemini 4 Argon features an output token limit of 1 million tokens, increased from the previous 64,000-token limit.
How much does Gemini 4 Argon cost to use via API?
Introductory pricing is set at $2 per million input tokens and $10 per million output tokens, with cached input discounted by 95%. Standard pricing will later be $4 per million input and $20 per million output tokens.
Who currently has access to Gemini 4 Argon?
Initial access is restricted to cybersecurity partners via Google's Fairwind Program and pre-release U.S. government vetting, with wider availability coming soon to paid API customers and Google AI Ultra subscribers.
How did Gemini 4 Argon perform on coding and security benchmarks?
Argon achieved 77.9% on DeepSWE v1.1 for software engineering and tied for first place on the CWE-bench v1 security benchmark with a 68% score.
Sources
- Gemini 4 Argon: our next era of frontier intelligenceGoogle · Sep 30, 2026 · Official
- Google rolls out new Gemini AI model but restricts access over safety concernsThe Guardian · Oct 1, 2026
- Google's First Gemini 4 Model Is 'Argon'Engadget · Oct 1, 2026
- Google announces Gemini 4 Argon as its new frontier model9to5Google · Sep 30, 2026
How this story was made: the newsroom picked it up from Google News, gathered the full text of the sources above, and drafted it with AI assistance. Every factual claim was then checked against those sources before publishing (21 claims checked). Illustrations marked as AI-generated are not photographs. Spotted an error? Tell us.
Published October 2, 2026 at 00:14 UTC
