According to CoinMeta, Gemini and Argon have released a new generation of flagship models, focusing on long-term complex tasks across software engineering, finance, legal work, and cybersecurity. The maximum output per instance of this model has been increased from the previous 64k tokens to 1 million tokens, allowing for continuous reasoning and execution of longer sequences of steps. The model is not yet available to general users; the first batch is only provided through Fairwind Program to trusted cybersecurity teams, with gradual access being opened to paying API customers and Google AI Ultra users in the future. In official evaluations, Argon achieved a score of 77.9% in Deepswe and v1.1, which is higher than 74.1% for GPT-6 and Astra and 74.2% for Claude Opus. Google has also enhanced Argon's cybersecurity capabilities, enabling it to automatically detect, verify, and fix software vulnerabilities. In the initial phase of its launch, API charged $2 per million inputs and $10 per output, with prices later doubling to $4 and $20 respectively.