Google's Next-Gen 'Gemini 4 Pro' Spotted in AI Arena
Google appears to be secretly testing its flagship 'Gemini 4 Pro' model, codenamed 'Argon,' under the guise of 'gemini-3.8-flash' on the AI Arena platform.
Image source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/
Google's upcoming flagship model, 'Gemini 4 Pro' (internal codename 'Argon'), has been spotted undergoing testing on platforms such as the LMSYS Chatbot Arena. It appears Google is quietly validating the model by masquerading it under the existing 'gemini-3.8-flash' identifier.
This discovery, circulated through community channels and major IT news outlets, highlights preliminary benchmark results where the model appears to significantly outperform current competitors, such as Anthropic’s Fable 5.1 and OpenAI’s Astra.
'Argon' Codename and Masquerading Strategy
According to analysis from technical experts and the developer community, the model's internal codename is 'Argon.' By utilizing the identifier of a previously released model, 'gemini-3.8-flash,' Google seems to be conducting field tests within the Chatbot Arena environment without immediately disclosing the flagship's identity.
While such strategies for gathering real-world data and validating performance before a major launch are not unprecedented, this instance is particularly notable as it involves the core model intended for Google's next major release.
Preliminary Benchmark Results and Outlook
Leaked data suggests that Gemini 4 Pro has achieved high benchmark scores across various metrics, including coding capabilities, complex reasoning, and agentic workflows. Its reported performance edge over models like Astra and Fable 5.1 suggests significant advancements in Google's AI development roadmap.
It is important to note that all current information is based on leaks and test environment observations; Google has not yet made an official announcement. Model naming and final specifications are subject to change upon official release.