Google has announced Gemini 4 Argon, a new frontier model built for complex, long-horizon professional workflows. Launched on September 30, 2026, the model targets software engineering, enterprise knowledge work, and cybersecurity defense. Argon can sustain deeper reasoning across extended tasks while raising the output limit to 1 million tokens, up from 64,000 tokens. Google says this capacity supports longer reasoning trajectories for difficult problems.
As stated by Google, “Gemini 4 Argon delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense.”
Notably, the model is partially available for cyber defenders as of now. Google will announce its general availability for developers, enterprises, and users in the near future.
Argon Moves into Google’s Internal Workflows
Argon is already being used inside Google while thousands of Googlers are testing it for specialized coding, research, and writing tasks. The company cited several internal projects to demonstrate its capabilities.
In one quantum computing project, Argon helped optimize resource use in computational subroutines. Google said the system beat a published baseline by 40% within minutes. Another deployment analyzed data center telemetry and identified memory optimizations.
Google estimated that these optimizations could free more than 300 TiB of memory after rollout while total savings could reach between 500 TiB and 1 PiB. Argon is supporting large codebase migrations from C++ to Rust for software engineering. The work spans projects from tens of thousands of lines to more than 800,000 lines. Google reported a 2.7x speed improvement for a Rust version of its libgav1 video decoder. The results followed extensive profile-guided experimentation and compiler analysis.
Performance Achieved Across Enterprise Tasks
Google’s model evaluation process tracks performance across coding, computer use, knowledge work, science, mathematics, multimodal tasks, long context, and cybersecurity.
Gemini 4 Argon achieved 77.9% on DeepSWE v1.1, a benchmark for long-horizon software engineering. It also reported a 51.3% score on AutomationBench. The model also performed strongly on visual and multimodal tasks. Google reported 91.7% on LVBench, which evaluates long-video understanding.
The evaluation methodology also notes that several comparison figures come from provider-reported results, while some Argon scores were independently computed by Google using specific evaluation harnesses.
Cybersecurity Is a Major Focus
Google trained Argon to find, validate, and patch critical software vulnerabilities. The company said trusted defenders will receive access without cyber guardrails. On CWE-bench v1, Google reported a 68% score for Argon. The result tied for first place on that benchmark.
Google also highlighted testing against prompt injection and other misuse risks. The company said Argon underwent automated and manual red-team testing. The initial rollout is limited while Google continues strengthening safeguards. Argon is being provided to trusted cyber defenders through the Fairwind Program.
According to Google, wider access will follow later, beginning with paid API customers and Google AI Ultra subscribers. The introductory API price is $2 per million input tokens and $10 per million output tokens.
WisdomPlexus brings the latest AI updates and announcements. So, stay in the loop with us and take a step toward a knowledge-driven tomorrow.
Read More : Google Gemini Image-to-Video with Veo 3: How to Turn Photos into Videos in 3 Steps
