Jan 11, 03:36 AM

Microsoft Launches Phi-4 Model, 14B Parameters, 7c/14c per Mtokens, Outperforms GPT-4o by 4.5%

Microsoft has launched its Phi-4 model, a 14 billion parameter AI, now available on Hugging Face at a competitive price of 7 cents per 14 cents per Mtokens. The Phi-4 model has demonstrated the ability to outperform larger models, including OpenAI's o1, by 4.5% in mathematical reasoning tasks. This release marks a significant step in enhancing the capabilities of small language models (SLMs) through Microsoft's 'rStar-Math' technique. Additionally, the Phi-4 model has achieved high scores on BigCodeBench, indicating its effectiveness among open-weight models. The model is also adaptable to the Llama architecture developed by Meta, allowing for more precise fine-tuning. Microsoft aims to further evaluate Phi-4's performance through human testing in various challenging scenarios.

#Microsoft #Hugging Face #OpenAI #BigCodeBench #Llama #Meta

Written with ChatGPT (GPT-4o mini).