A user compares local quantized Qwen 3.6 models against frontier models on a single-file HTML canvas driving animation task, finding that the local 27B Qwen quant delivers competitive results with better parallax and motion than some frontier outputs.
Saw [this post](https://www.reddit.com/r/LocalLLaMA/comments/1styxdy/compared_qwen_36_35b_with_qwen_36_27b_for_coding/) comparing Qwen 3.6 variants on coding primitives, so I wanted to see how local quants stack up against frontier models on a similar dense, single-file coding task. I ran the exact same prompt across local and web-based models accessed through my Perplexity subscription. The prompt "Write a single HTML file with a full-page canvas and no libraries. Simulate a realistic side-view of a moving car as the main subject. Keep the car visible in the foreground while the background landscape scrolls continuously to create the feeling that the car is driving forward. Use layered scenery for depth: nearby ground, roadside elements, trees, poles, and distant hills or mountains should move at different speeds for a natural parallax effect. Animate the wheels spinning realistically and add subtle body motion so the car feels connected to the road. Let the environment pass smoothly behind it, with repeating but varied scenery that makes the movement feel believable. Use cinematic lighting and a cohesive sky, such as sunset, dusk, or daylight, to enhance atmosphere. The overall motion should feel calm, immersive, and realistic, with a seamless looping animation." **Models tested** Frontier (web-based via Perplexity, tok/s not measured): * Claude sonnet 4.6 Thinking — used internet for reasoning * Gemini 3.1 Pro Thinking * GPT 5.4 Thinking * Kimi k2.6 Thinking Local (Ryzen 5 5600, 24 GB DDR4-3200, RX 5700 XT 8GB): * Qwen3.5 9B Q4\_K\_M — \~50 tok/s * Qwen3.6-27B (Claude-opus-reasoning-distilled) Q4\_K\_M — 2.65 tok/s * Qwen3.6-27B Q4\_K\_M — 2.70 tok/s * Qwen3.6-31B A3B Q4\_K\_M — 12.13 tok/s * Gemma-4-31b-it — 1.91 tok/s * Qwen3.5 4B Q8 — 60 tok/s — used internet for reasoning * Qwen3.5 4B Q4\_K\_M — 80 tok/s — used internet for reasoning **What I looked for** Realistic side-view driving animation: layered parallax scenery, spinning wheels, subtle chassis motion, cohesive sky and lighting, and seamless looping — all vanilla JS/canvas, zero libraries. **Subjective ranking for this specific task** 1. Kimi k2.6 Thinking — cleanest overall visual result 2. Qwen3.6-27B Q4\_K\_M (local) — stronger than I expected; good parallax and road feel 3. Qwen3.6-27B Claude-opus-reasoning-distilled — close third The local 27B quant delivered more natural motion and layering than some frontier outputs for this specific visual primitive. I was expecting frontier models to do much better — am I missing something? **Outputs** I only changed the HTML `<title>` tags to track which model generated which file. I’ll share all the output files and probably a few screenshots of the running animations so you can judge the visual quality yourself. If anyone wants to run the exact same prompt on their setup — especially other MoE cuts or distills — feel free to share your results.
A user benchmark demonstrates that the Qwen 3.6 27B dense model (Q4 quantized) can autonomously generate a fully playable multi-file game in a single prompt on a single RTX 3090, significantly outperforming its predecessor with zero manual interventions. The results highlight major improvements in local code generation and agentic capabilities for consumer-grade hardware.
A user compares Qwen3.6 35B-A3B and Gemma 4 26B-A4B-IT running locally on a 16GB VRAM GPU via LM Studio, finding Qwen3.6 produces more detailed outputs while both run at comparable speeds. The post is an informal community comparison using quantized models.
An informal benchmark comparing 8 AI models (Qwen3.6 35B, Qwen3.5 series, Gemma 4 series, GLM 4.7 Flash) in creating racing games via OpenCode/Playwright MCP, testing their coding agent capabilities and documenting various implementation quirks.
The author highlights the impressive capabilities of the open-source Qwen 3.6-27B model running locally on an RTX 5090, noting its strong performance on programming tasks and comparing it favorably to commercial models, despite the complexity of local deployment.