I tested 9 LLMs on the exact same web-dev prompt for ~8 hours — RTX 3060 12GB results (Rate the best!)

A developer spent approximately 8 hours testing 9 different Large Language Models (LLMs) on a single web-development prompt using an RTX 3060 12GB. The tested models included Gemini 3.8 Flash, GPT-5.6 Sol, Claude Sonnet 5, Bonsai 2 27B Ternary, Qwen 3.8 27B (various configurations), Ornith 1 9B Q4_K_M, and Ornith 1.5 35 A3B Q6. Local models like Bonsai 2 and Ornith 1 showed generation speeds of ~34–36 tok/s and ~74 tok/s, respectively. The developer also requested results for GPT-6-ASTRA.