Model Evaluation

Qwen3.6-27B — UD-Q5_K_XL

Agentic reasoning, production-grade front-end design, and canvas / WebGL creative coding. 19 runs, 93.9k tokens generated, self-hosted on a single RTX 5090.

Read the full report Round 2 · Speed + DFlash Follow @KyleHessling1

55.3avg tok/s

19runs

93,899completion tokens

22.1 GBVRAM used

65Kcontext (q8 KV)

Web design · open to preview

SaaS landing pagePrism — AI observability

35.8 KB · 9,991 tok · 180 s

Analytics dashboardLight theme, emerald accent

40.8 KB · 12,733 tok · 230 s

Designer portfolioMaya Chen — kinetic typography

20.9 KB · 5,387 tok · 97 s

Pricing page3 tiers + animated toggle + FAQ

29.2 KB · 7,849 tok · 141 s

Mobile app marketingStillwater — CSS-only iPhone mock

32.4 KB · 9,243 tok · 166 s

Canvas / WebGL · creative coding

Particle attractor3000-particle fluid swarm

13.1 KB · 4,604 tok · 82 s

WebGL MandelbulbRaymarched fractal shader

15.2 KB · 4,940 tok · 88 s

Three.js crystal sceneTransmissive glass + bloom

19.9 KB · 6,490 tok · 116 s

Physics sandboxSoft-body collisions, fling mouse

21.2 KB · 7,302 tok · 131 s

Audio-reactive visualizerMic + oscillator fallback

17.8 KB · 6,432 tok · 115 s

Agentic reasoning · text output

Multi-step planningURL shortener deploy plan

thinking: 3,802 tok · 68 s

Self-critique loopPalindrome · O(n³) → O(n²)

thinking: 2,837 tok · 51 s

Code debug (4 bugs)k-th smallest element

no-think: 263 tok · 5 s

Structured JSON extractionCalendar + roster from prose

no-think: 250 tok · 5 s

Tool-use planningWeather + flights + hotel

no-think: 211 tok · 4 s