"MiMo-V2.5 claims to be a major step forward in agentic capability and multimodal understanding, achieving best-in-class performance on its in-house agentic tasks benchmark."
"The smaller V2.5 model matched the larger V2.5-Pro at half the cost, demonstrating significant efficiency in performance across various benchmarks."
"Trained on 48 trillion tokens, MiMo-V2.5 is natively multimodal, supporting text, image, and video data, with a context capacity of 1 million tokens."
"Users can download MiMo-V2.5 from Hugging Face or try it in the AI Studio, but high-performance hardware is necessary for local execution."
Xiaomi has launched MiMo-V2.5, an open-weight AI model that excels in agentic capability and multimodal understanding. It achieved best-in-class performance on in-house benchmarks, matching the larger V2.5-Pro model at a lower cost. Trained on 48 trillion tokens, MiMo-V2.5 supports text, image, and video data, with two versions available: a 310B parameter model and a 1.02T parameter Pro version. Users can download the model from Hugging Face or access it via an API, although high-performance hardware is required for local execution.
Read at GSMArena.com
Unable to calculate read time
Collection
[
|
...
]