Emir BardakçıXiaomi has rolled open the new ChatGPT-style web AI platform, which is fueled by the newly unleashed MiMo-V2-Flash AI Model. Such a platform enables consumers to experience firsthand Xiaomi’s latest large language AI without needing to install any software. This move signals one of the biggest breakthroughs offered by Xiaomi in the AI domain. Xiaomi had generally targeted the smartphone market until now.
Xiaomi Launches MiMO Studio for Ai Chat Interactions
“Xiaomi MiMO Studio,” launched recently, provides an interactive online AI communication platform for trying out “MiMo-V2-Flash,” an AI model. It has options for both “deep reasoning conversations” and “online search enhanced responses,” making it useful for coding, research, or general knowledge. To try it yourself just go to aistudio.xiaomimimo.com.
Initially, Xiaomi has revealed that the demand for the devices has been well above expectations. Due to this, there could be a delay in accessing the servers or a call to retry the login process owing to overload traffic. Initially, there are no limitations announced for specific geographic areas.
Model for MiMo-V2-Flash Designed for Speed and Efficiency
MiMo-V2-Flash is an in-house “Mixture-of-Experts (MoE) model” designed exclusively for the scenario of an intelligent agent and is a large language model developed by Xiaomi. The model includes “309 billion total parameters” and “15 billion active parameters” for inference.
This model adopts Hybrid Attention architecture, incorporating both Global Attention and Sliding Window Attention (SWA) with a 1:5 split. It natively trains on a 32K tokens context with scalability up to 256K tokens, and MiMo-V2-Flash is specifically trained to excel in reasoning tasks, code generation tasks involving intense processing, and multi-step agents.
Open-Source AI Competitive Performance
As asserted by Xiaomi’s technical revelations, MiMo-V2-Flash is one of the top two open-source AI models in the world in terms of several general-agent targeted benchmarks. The programming and logic skills of MiMo-V2-Flash outpace all other open-source models and are of similar caliber to closed-source models.
Nevertheless, in spite of its performance, Xiaomi claims a reduction in the inference cost to around 2.5% of comparable closed models, while its generation speed is accelerated by two times. This makes MiMo-V2-Flash a very suitable approach for developers and enterprises who require scalable AI without incurring high operational costs.
Open Source, API Pricing, and Developer Access
The MIT license was adopted for the release of the model weights and inference code, further showing the company’s dedication to open-source software. MiMo-V2-Flash can also be used with API access by the developers at the following prices:
- $0.10 per million input tokens
- Cost of $0.30 per million tokens
There is a limited period for which the usage of APIs is offered free of cost. This facilitates the testing and integration of APIs into third-party apps.
