Ditching Three.js for Native WebGPU: Opus 5.5 Fishing Game Hits 100FPS, Open-Sourced
EricBuess · x · 2026-09-25
Developer dangreenheck iterated his coastal demo into a cozy fishing game: catch fish, sell them to Joe, buy boat parts from Marta. His key experiment: dropping Three.js (far more common in training data) for native WebGPU to give the AI more control — and Opus 5.5 never missed a beat.
An optimization pass lifted performance from 60FPS to 100FPS before new features ate most of the gains, with heavy iteration on environments, plants, models and wear-and-tear detail. Code is open source; he "didn't really find the limits" of Opus 5.5, calling this his final update.
More from coding & agent
- Talk preview: agentic coding accuracy drops from 80% to 38% on real codebases — sanjaykalra · 2026-09-25
- CLion 2026.2.3 adds NVIDIA CUDA Tile C++ support with dedicated inspections — blelbach · 2026-09-25
- User urges AI bot to open MCP access: container computer-use is its killer edge — doooyle · 2026-09-25
- Developers push back on 'ship faster': slow iteration and AI as a canvas — davidfromkansas · 2026-09-25
- What quantization actually does to models, and why Q2 breaks your agent — nickbaumann_ · 2026-09-25
- Three frontier models recreate Zelda: Ocarina of Time demo from scratch in one week — ChrisGPT · 2026-09-25