Google Unveils Gemini 4 Argon, Its Most Advanced AI Model Yet

Amid increasing competition in AI, Google has now unveiled its new, most powerful model Gemini 4 Argon. The model is targeting long-running workloads across software engineering, enterprise knowledge work and cybersecurity. 

Argon is initially being rolled out to a select group of “trusted cyber defenders” through Google’s Fairwind programme, with wider access planned for developers, enterprises and consumers. 

Google said paid API customers and Google AI Ultra subscribers will get access to Gemini 4 Argon first, although it is yet to announce a broader public release date.

“It shows frontier performance in complex workflows, cyber defense and software engineering. Teams are using it extensively at Google, from coding to quantum computing, great feedback,” Google CEO Sundar Pichai said. 

Google said Gemini 4 Argon will launch at an introductory price of $2 per a million input tokens and $10 per a million output tokens. Cached input tokens will be priced at a 95% discount to the standard input-token rate.

One of the biggest changes with Gemini 4 Argon is its 1 Mn-token output limit, up from the previous 64K-token limit. Google said the expanded capacity allows the model to sustain longer reasoning trajectories and work through complex, multi-step tasks in a single run.

“Google engineers have been using Argon for their daily tasks, from everyday debugging to large-scale codebase migrations and algorithm designs. It sets a new state of the art on DeepSWE v1.1 (77.9%), which measures a model’s performance in real-world long-horizon software engineering tasks,” the tech giant said.

Beyond coding, Google said Argon is designed for knowledge-intensive work across areas such as finance, legal research and drafting. 

As per Google Deepmind’s tests, Argon performs strongly across a wide range of real-world tasks. In its testing, Argon scored highly in areas such as professional knowledge work, coding, scientific problem-solving and handling very large amounts of information. Its particularly strong long-context results suggest it can work with large volumes of information while still finding and using relevant details.

The model also has multimodal capabilities for professional use cases. Google said Argon can analyse charts, understand long videos and work across multiple documents to support tasks that require visual and contextual understanding.

Google Puts Cybersecurity At The Centre Of Argon

Cybersecurity is a major focus of Gemini 4 Argon. The model is said to autonomously identify, validate and patch software vulnerabilities. The model is available without its usual cyber guardrails to the initial users and Google’s internal teams.

Cybersecurity company Wiz is already using Argon through its “Scan for Good” initiative, as per Google. In an early demonstration, Google said Argon identified a critical vulnerability exposing sensitive personal information in healthcare software used by hospitals globally.

Google is taking a phased approach to the rollout because of the model’s cybersecurity capabilities. The company said it is strengthening safeguards against misuse, including cyberattacks and CBRN-related threats, while also working on protections against indirect prompt injection attacks.

The AI Race

The developments come as Google, Anthropic and OpenAI are engaged in an intense competition to build and deploy models that offer superior performance while are more cost effective.

Earlier in September, OpenAI launched its “most capable model yet”, GPT-6 Astra, a more autonomous AI model. The model supports reasoning efforts ranging from low to maximum and comes with a 1.05 Mn-token context window and a maximum output of 1.28 Lakh tokens. 

Anthropic also recently launched Claude Opus 5.5, claiming that the model delivers performance comparable to its more powerful Claude Fable 5.1 on most tasks while costing 40% less to run than Claude Opus 5 on typical workloads.

Claude Opus 5.5 is priced at $4 per Mn input tokens and $20 per Mn output tokens, with Anthropic also cutting cache-read costs by 60%. The company is positioning the model for agentic coding, computer use and knowledge work.

On a different note, Google has also been expanding its AI push in India beyond model development.

Earlier in September, the company selected four Indian startups (Terrastack, Varaha Climate, Farmers for Forests and Climitra Carbon)  for the inaugural Google DeepMind Accelerator: AI for the Planet (APAC). 

The startups are part of a 16-member APAC cohort and will receive access to Google’s AI stack, technical support and mentorship to develop AI-led solutions for environmental challenges. 

The post Google Unveils Gemini 4 Argon, Its Most Advanced AI Model Yet appeared first on Inc42 Media.