Qwen3.8 27b Q4_K_XL impersonating a frontier model on this HTML tutorial build.
Kimi k3, Grok4.6 & DeepSeek 4pro, all given the same 3 svg’s, 26kb text file and 17kb prompt. API cost: Kimi $2.30, Grok 48c, DeepSeek 36c, Qwen 🍩 Kimi the nicest, Grok clean and cheap, Qwen 27 tho🤯
[Shocking] Chinese AI has reached the top level in the world! The real reason why Japan is dangerous ✎︎＿＿＿＿＿＿＿＿＿
Are you avoiding Chinese AI because it's suspicious? In fact, China is gaining hegemony behind the latest AI being used around the world. ✎︎＿＿＿＿＿＿＿＿＿ ■“The evolution of Chinese AI and the facts we should know” 1.
AI reads your prompt once, front to back.
Pasting the same prompt a second time underneath improves the answer. The December 2025 paper Prompt Repetition Improves Non-Reasoning LLMs tested seven models from the Gemini, GPT, Claude and DeepSeek families, told not to reason first.
"Draw a 5-a-side soccer court" - This is the story of an experiment in which when the same instructions…
You have reached the end of the archive
All of deepseek