Gemini 3.0 spotted in the wild through A/B testing (ricklamers.io)

A user reports spotting what appears to be Gemini 3.0 in the wild via an A/B test inside Google AI Studio. Using a simple structured prompt—“Create an SVG image of an Xbox 360 controller. Output it in a Markdown multi-line code block.”—they received a notably higher-quality SVG (model ID ecpt50a2y6mpgkcn) that outperformed the competing frontier model visually. The community has been using SVG tasks (e.g., Simon Willison’s “pelican riding a bicycle” test) as an efficient proxy for coding/structured-output quality, and this example reinforced that proxy’s usefulness for quick, human-evaluable comparisons. Technically, the observed variant produced about 24 seconds higher time-to-first-token and roughly 40% longer outputs (including apparent reasoning tokens), which suggests more extensive internal generation but not necessarily massive test-time compute like a hypothetical “GPT-5 Pro.” The test-selection and opaque model ID leave open whether this is Gemini 3.0 Flash vs. Gemini 3.0 Pro or a 3.0 vs 2.5 matchup (the user had selected Gemini 2.5 Pro), so conclusions are preliminary. Still, the A/B sighting is a meaningful early signal for the AI/ML community: Google may be quietly rolling out a model with measurable gains in code-like and structured generation tasks, which could presage improvements in coding assistants and deterministic-output applications.