Eseandre Mordi
⬤ Xiaomi just dropped MiMo-V2-Flash, a 309-billion-parameter open-weights reasoning model that landed a 66 score on the Artificial Analysis Intelligence Index. The release puts Xiaomi in the same league as major AI labs, showing the company's serious push into large-scale AI development.
⬤ The model really shines in tool-use and math competitions. MiMo-V2-Flash crushed it with 95% on τ²-Bench Telecom and 96% on AIME 2025, taking the top spot in the τ²-Bench Telecom category. These numbers show the model's solid chops at multi-step reasoning and working with complex tools.
⬤ What's grabbing attention is the price tag. Running the entire benchmark suite cost just $53, with pricing at $0.10 per million input tokens and $0.30 per million output tokens. That puts it close to DeepSeek V3.2 in cost and way below GPT-5.2, which needed significantly more cash to benchmark. The model does burn through tokens though, using roughly 150 million reasoning tokens across the full evaluation.
⬤ Xiaomi released MiMo-V2-Flash under an MIT license with 309B total parameters and 15B active during inference. It's another example of Chinese AI labs going all-in on open-source frontier models, mixing competitive performance with aggressive pricing that works for budget-conscious deployments.
Eseandre Mordi
