Is Flash really at Opus level? Eight builds, four prompts
People say Google's Flash series jumped from roughly GPT-5 level to Opus level in a few months. So the same four prompts, word for word, went to Claude Opus 5.5 and Gemini 3.8 Flash: a floatplane simulator (the test reviewers used on the leaked Gemini 4 checkpoints), a village of NPCs that remember you, a photo editor and a swarm of bug-hunting agents. One request each. Everything runs in your browser.
A floatplane simulator
The test reviewers used on the leaked Gemini 4 checkpoints: a 3D floatplane that taxis on water and takes off.
It plows, climbs onto its floats with spray flying, lifts off, and a STALL warning lights up if you pull too hard. Reasoning low. No edits.
Nice islands, but on the water it reaches 150 knots, never takes off cleanly and the water is one hypnotic ripple. No edits.
A village that remembers you
Six villagers with routines, memory and gossip: the idea behind AI NPCs in games.
Bright and cozy, with a fountain, a baker, a smith and a journal. Talk to Marta twice and she tells you that you already said that today. Reasoning low (the first try was cut by a 64k output limit and run again with 110k). No edits.
A darker sunset village, but with a ledger of what each villager remembers about you and the gossip they have heard from the others, plus two quests. No edits.
A photo editor
A pro-style editor in a single page: layers, curves, histogram, presets, undo, export.
Layers with blend modes, a live histogram, curves, crop, brushes, text and export. Press Sample to paint the demo photo. Reasoning low. No edits.
Adjust, presets, layers and history tabs, curves and a histogram. Press Sample Image. No edits.
A swarm of bug hunters
A command centre for hundreds of AI agents hunting 70 planted bugs in a 400-file codebase.
It works: the chart climbs to 69 of 70. But the map of the codebase is a thumbnail in the corner and most of the page is an empty chart. Reasoning low. No edits.
A huge map where every file is a tile, agents swarming over it, bugs flashing red then green, a live feed and a swarm-versus-single-agent benchmark. No edits.
The exact prompts are for Ruocco Academy members. Every prompt word for word, with model, settings, time and cost, the full clean source of every demo as a ZIP, and step-by-step notes on how each video was made. The demos on this page are free to use for everyone.
Join the Academy