Mimo V2.6 Pro & Flash (Fully Tested): This is OPEN WEIGHTS!?
In this video, I compare Xiaomi’s new MiMo V2.6 Pro and MiMo V2.6 Flash models across all eight KingBench 3 tasks using OpenCode and OpenRouter. I test their coding, 3D generation, game development, SVG design, math reasoning, local fine-tuning, and overall reliability to find out which model offers the best performance and value. -- Key Takeaways: 🚀 MiMo V2.6 Pro and Flash are open-weight mixture-of-experts models with context windows of over one million tokens. 💸 MiMo V2.6 Flash is about 68 percent cheaper than Pro for standard input and output tokens. 🏢 Flash produced a more reliable elevator simulation, while Pro suffered from passenger-counting and reset bugs. 🎨 Flash created the better panda SVG, although neither model fully captured the panda taking a bite. 🎮 Both models built functional bow-and-arrow games, but they shared an input-handling bug involving the R key. 🧮 Neither model completed the original math task, though Flash made substantial progress during a larger-budget retest. 🛠️ Both models successfully completed the local Gemma 2B fine-tuning task and earned full marks. ⌚ Pro delivered more accurate watch hands, while Flash’s analog time calculations contained major errors. 🏆 Flash finished with the higher overall score and appears to be the better value for inexpensive prototypes.
