MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis

Open-weight models now offer top-tier intelligence, challenging proprietary systems on key benchmarks.

Xiaomi has released MiMo-V2.6-Pro, an open-weights model evaluated on the Artificial Analysis Intelligence Index. The model scored 46, significantly above the median of 18 for comparable models. It supports text, image, speech, and video inputs with a 1M token context window. MiMo-V2.6-Pro excels in intelligence and speed, outputting 124.5 tokens per second, placing it 12th out of 114 models in speed. However, it is noted as somewhat verbose. In terms of cost, input tokens are priced at $0.43 per 1M and output tokens at $0.87 per 1M, with an evaluation cost of $206.66 for the Intelligence Index tasks. The model has 1.0T total parameters and 42B active parameters, licensed under MIT and available on Hugging Face.

MiMo-V2.6-Pro's strong performance, especially in intelligence and speed, demonstrates the rapid advancement of open-weight models. Its ability to handle multimodal inputs and a large context window positions it as a versatile tool, potentially driving innovation and competition in the AI landscape. This release suggests a democratization of advanced AI capabilities, making powerful tools more accessible to a wider range of developers and researchers.

The pricing model for MiMo-V2.6-Pro, while noted as somewhat expensive for input tokens, offers a competitive alternative to proprietary models. The availability of such high-performing open-weight models with strong benchmarks may pressure cloud providers and AI service companies to adjust their pricing and performance offerings. Developers will need to weigh the costs against the flexibility and advanced capabilities offered by models like MiMo-V2.6-Pro.