Google DeepMind has introduced its new frontier model, Gemini 4 Argon, which achieved a score of 53 points on the Artificial Analysis Intelligence Index. This performance places the model among the top three AI labs, surpassing OpenAI’s GPT-6.1 Sol by one point and matching GPT-6 Astra.
The model is designed for complex tasks, including coding, cybersecurity, and enterprise knowledge management. It features a 1 million token output limit, a significant increase from the previous 64 000 token limit. In testing, Argon demonstrated a 15% hallucination rate compared to 51% for Astra, though it scored 13 points lower in accuracy on specific questions.
Google has launched the model under the Fairwind Program, currently restricting access to a group of trusted cyber defenders and US government entities. Internal Google teams are also utilizing the model for tasks ranging from software development to quantum computing research.
Introductory pricing is set at $2 per million input tokens and $10 per million output tokens, which will double once the promotion concludes. At current discounted rates, the cost per index task is $1.99, compared to $3.26 for Astra. However, at standard rates, the cost per task increases to $3.98.
Argon demonstrated strong agentic capabilities, leading the AutomationBench-AA with a 77.5% score, ahead of Anthropic’s Claude Sonnet 5.5. Future access will be granted to paid API customers and Google AI Ultra subscribers, though details regarding the duration of the introductory pricing remain unconfirmed.
