Xiaomi MiMo-V3 Pro and Flash reportedly rival DeepSeek V4.1-Flash

Open-weight models keep getting cheaper and stronger. The next push may not come from an AI lab. Leaks say Xiaomi is preparing two models aimed straight at DeepSeek.

No official word yet, but the leaked details are specific enough to watch. If they hold, the cost and capability balance in coding and agent tasks could shift.

What Happened

Developer chatter ahead of DeepSeek's stealth V4.1-Flash test window points to Xiaomi's counter-move: MiMo-V3-Pro and MiMo-V3-Flash. The flagship V3-Pro reportedly scores about 72.8 on SWE-Bench Pro to take on closed-source titans. The V3-Flash is described as an ultra-lean tier built for low-latency agent tasks.

DeepSeek's answer is V4.1-Flash, designed to hold its lead on very low-cost inference while adding native multimodal agent speeds without giving up core reasoning.

Why It Matters

Xiaomi is aggressively stacking its ecosystem — smartphones, EVs, IoT — with in-house foundation models. That makes this a hardware giant versus an open-weight disruptor. If the leaks are real, device makers could give dedicated AI labs real competition on cost and local performance in everyday coding and agent work.