Xiaomi has entered the open-weight AI race with MiMo-V2.6-Pro, which has taken the top position among open-weight models on Artificial Analysis’ Intelligence Index.
The phone and electric car maker has released 2 open-weight models, Pro and Flash, under an MIT licence, with their weights available on Hugging Face. Artificial Analysis scores MiMo-V2.6-Pro at 46, placing it 1st among 114 models in its category and making it the highest-scoring open-weight model on the index.
Xiaomi says Pro was trained in under 6 days for around $2.62 million. It is a sparse mixture-of-experts model with 1.02 trillion total parameters, of which 42 billion are active per token. The model supports text, images, audio and video and has a 1 million-token context window.
The company charges $0.435 per million input tokens and $0.87 per million output tokens. Xiaomi says Pro outperforms Moonshot’s Kimi K3 and Alibaba’s Qwen3.8 Max.
Both Pro and Flash are available through Xiaomi AI Studio, MiMo Code, MiMo Desktop, Xiaomi’s API platform and OpenRouter. Developers can also download the weights from Hugging Face for self-hosting.
A faster version, MiMo-V2.6-Pro-UltraSpeed, can generate up to 20 times faster than Pro at the same quality, according to Xiaomi. It costs $4.35 per million input tokens and $8.70 per million output tokens.
Pro and Flash completed 30 training steps across around 750,000 trajectories. Flash cost around $850,000 to train and is priced at $0.14 per million input tokens and $0.28 per million output tokens.
Xiaomi reports that Pro improved its DeepSWE software engineering score from 58.4 to 72.57 during training, while Flash rose from 48.8 to 65.68. Flash also surpassed Pro on the CyberGym cybersecurity test, scoring 95.1 versus 94.0.
The company says it trained coding, general agents, visual tasks and cybersecurity together rather than using separate training runs. It also introduced measures against reward hacking after finding agents exploiting future fixes instead of solving assigned bugs.
Alongside the models, Xiaomi published its technical report, reinforcement learning framework, more than 7,000 task environments, automatic graders, reward design, hyperparameters, data mixtures and training costs.
Artificial Analysis still places models such as Claude Opus 5.5 above MiMo-V2.6-Pro overall. Xiaomi’s advantage is its open-weight position, lower operating costs and the amount of training material it has made available for researchers.
Also read: Viksit Workforce for a Viksit Bharat
Do Follow: The Mainstream LinkedIn | The Mainstream Facebook | The Mainstream Youtube | The Mainstream Twitter
About us:
The Mainstream is a premier platform delivering the latest updates and informed perspectives across the technology business and cyber landscape. Built on research-driven, thought leadership and original intellectual property, The Mainstream also curates summits & conferences that convene decision makers to explore how technology reshapes industries and leadership. With a growing presence in India and globally across the Middle East, Africa, ASEAN, the USA, the UK and Australia, The Mainstream carries a vision to bring the latest happenings and insights to 8.2 billion people and to place technology at the centre of conversation for leaders navigating the future.