In recent months, OpenAI and Anthropic pushed flagship models forward while Google stayed quiet, updating Flash roughly every three weeks with 3.6, 3.7, and 3.8 iterations but showing no progress on the Pro variant. A model labeled gemini-3.8-flash appeared in Arena, the large model blind test arena, delivering noticeably higher performance in writing code, generating SVG, and running Agent tasks than the official Gemini 3.8 Flash, which Google released on September 2 with significant improvements in software engineering, Agent tasks and complex multi-step reasoning, only three weeks after 3.7 Flash. After several rounds of testing, speculation spread that the model is likely Gemini 4 Pro. A benchmark comparison chart went viral showing it outperformed GPT-6 Astra and Claude Fable in coding, Agent operation, and reasoning. Developer Harshith compared its SVG cat side view with GPT-6 Astra Max and official Gemini 3.8 Flash, finding visible gaps in outline, proportion and detail completeness. X user @thtbee_ tested it generating a Voxel-style pagoda in 8 minutes as a full set of interactive 3D scenes. Behind the suspected Gemini 4 Pro is the keyword RSI.