Microsoft reveals cutting-edge MAI-Base-2 model with 2T parameters
Microsoft Corp. has developed a new large language model that it says is significantly more efficient than leading open-source alternatives. The company’s artificial intelligence research group detailed the algorithm on Thursday. MAI-Base-2 is not a full-fledged LLM but rather a base model, a kind of no-frills prototype. Developers turn base models into LLMs by putting […] The post Microsoft…
Microsoft unveiled a new large language model called MAI-Base-2, boasting 2 trillion parameters and significantly outperforming existing open-source alternatives. The model, developed by Microsoft's artificial intelligence research group, is a base model that developers can transform into a full-fledged language model by undergoing additional training rounds focused on reasoning features and safety guardrails.
MAI-Base-2 excels in science, programming, and math questions, outperforming the most capable open-source base model, DeepSeek-V4-Pro-Base. Moreover, it is up to 12 times more hardware-efficient than its counterparts. The model utilizes a mixture-of-experts (MoE) architecture, using only a subset of artificial neurons to process prompts, making it more efficient than activating the entire neural network.
MAI-Base-2 features 80 billion activated neurons per prompt. Microsoft trained the model for 38 days using 16,000 Nvidia GB200 chips and a training dataset of 50 trillion tokens, with about a fifth of the tokens dedicated to a computer vision module that enhances its scientific chart understanding capabilities. Microsoft will license MAI-Base-2's weights to trusted partners through Azure and plans to incorporate the model into a new series of LLMs, with further details to be shared in the future.
Written by urgent.news from SiliconANGLE's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
