Over the past several months, OpenAI and Anthropic have launched their flagship models one after another, while Google has maintained a relatively low profile. Its Flash - series models have undergone updates nearly every three weeks, progressing from version 3.6 to 3.7 and then to 3.8. However, there has been a notable lack of new developments in its Pro - series, which stands as the pinnacle of its capabilities.

Until recently, a model named gemini - 3.8 - flash suddenly made an appearance in the large - model blind - test arena called Arena. It demonstrated significantly superior performance compared to the officially released Gemini 3.8 Flash. This has fueled speculation that it might be Google's next - generation flagship model, Gemini 4 Pro.

According to tests conducted by developers, this model excels in code generation, SVG creation, and Agent operation. In some tests, it even outperformed GPT - 6 Astra and Claude Fable.

In the meantime, it has been reported that Gemini 4 Pro, with the internal codename argon, supports output tokens of up to 256K and a context window of 2M. It also boasts advanced features such as cross - session permanent memory and internet connectivity without the need for an API. Although Google has not officially confirmed these details, rumors about Gemini 4 Pro have become closely intertwined with the Recursive Self - Improvement (RSI) technology that Google is currently exploring. RSI technology aims to empower AI to participate in the creation of even more powerful AI, thereby accelerating the evolution of the model itself.