Deepseek v4.1 1017 - Horse

I performed the usual horse benchmark to compare. The model worked and worked. It consumed the most amount of tokens I've ever experienced with this prompt. It used 10M tokens creating cloud and at the end they're invisible. It had access to Playwright for taking screenshots. It was continuously making edits and adjustments. I struggled to control myself not to stop it. web.esuyo.com/horse-deepseek-v4.1-1017 The visuals is one of the best out of all the models I've tested. The animation of the legs I'd hoped would have been better. It seemed to most of the time editing to make it visually attractive and not enough on the horse animation. I would have liked to have had pan and zoom controls like other provide. Statistics 419,551 tokens 40% used Session Token Summary — ↻ 259 turns Metric Value ↑ in 3.4M ↓ out 133.6K ≣ cache write 0 ≣ cache read 68.2M ∅ cache hit rate 95.3% ♦ think 233.9K Σ total 71.9M ◫ tools — 457 · 2 failed ↻ initiated — 1h47 → busy — 1h43 · 24s/step Prompt: Create a low-poly 3D horse running in a seamless loop using Three.js. Build it procedurally from primitive geometry, animate the legs, body, head, and tail. Organise the implementation into a clean, maintainable multi-file structure. Separate HTML, CSS, and JavaScript into appropriate files, and split JavaScript into multiple ES modules where doing so improves separation of concerns. The project must be self-contained and runnable by opening index.html (or using a simple static web server), with no build process required. External CDN dependencies such as Three.js are acceptable. Do your best, you are being rated against other AI models. Put extra effort in. The other horse posts for comparison: