@sky.wav: Today's hands-on test using the same prompt on Gemini 3.7 Flash vs Gemini 3.1 Pro, with uniform High intensity: 3.7 Flash: 216.1 seconds / 76 KB 3.1 Pro: 189.8 seconds / 45 KB In this round, 3.1 Pro is about 12% faster, while 3.7 Flash outputs about 71% more code. But code volume doesn't represent quality—the key is still to see if there are collisions or deadlocks under high traffic flow, and whether the statistical data is authentic. Personally, I think 3.7 Flash has higher completeness. Test prompt: Create a 3D intelligent intersection traffic scheduling simulation with a production-quality feel, adapted to a 1280×800 screen. Use an isometric or slightly overhead camera to fully display four-way roads, lane markings, crosswalks, traffic lights, street lamps, buildings, and greenery. Cars, buses, and pedestrians continuously generate randomly, adhering to lane rules, turning, traffic signals, and crosswalk rules, without penetrating each other. The interface must provide two traffic light modes: fixed cycle and AI adaptive; the adaptive mode dynamically adjusts green lights based on queue lengths in each direction and pedestrian waiting conditions, while setting minimum green light duration, yellow light transitions, and interlocking for conflicting directions to avoid collisions and permanent deadlocks. Real-time display of average wait time, cumulative throughput, current road vehicle count, and congestion index, and provide sliders for traffic density, mode switching, pause, reset, and day/night toggle. Statistical data must be genuinely calculated from the simulation state, not just random numbers. Visually include vehicle brake lights, glowing traffic lights, soft shadows, and smooth camera movement, default to auto-run without needing to click start. Only allow HTML, CSS, JavaScript, and local Three.js; do not load other libraries or network resources. The first draft must resemble a fully demonstrable complete traffic digital twin product. #gemini37 #aicode #aitest #aimodels #aifypシ

sky

Open In TikTok:

Region: VN

Thursday 20 August 2026 23:10:00 GMT