A new model named "gemini-3.8-flash" has quietly debuted on the large-model arena, and developers testing it believe it is likely Google DeepMind's next-generation flagship, Gemini 4 Pro. According to a viral benchmark image, Gemini 4 Pro outclasses GPT-6 Astra and Claude Fable 5.1 across coding, agent capabilities, and reasoning. Developer Vidhi was blown away after experiencing it, and an SVG image of a cat reportedly showed GPT-6 Astra producing a "big cat." Google has implemented RSI internally, as rumored, enabling Gemini 4 Pro to compete with Astra and Fable. Google's last major version update, Gemini 3.1 Pro, was seven months ago, on February 19. At Google I/O, Pi Chai announced Gemini 3.5 Pro would launch in June, but it was postponed and ultimately scrapped; earlier this month, the WSJ exclusively reported the cancellation because the 3.5 Pro's evolution didn't match Flash. Gemini 4 showed strong pre-training performance and smooth post-training. The Gemini 3.8 Flash name was unveiled as early as early September. Leaked benchmark screenshots show DeepSWE v1.1: 88%, nearly 2 percentage points higher than Astra; GDPval-AA v2: 2064 Elo, the only one to achieve it; Terminal-bench 2.1: 95.3%; OSWorld-