Home ยป Enhanced Efficiency: DeepSeek Unveils New V3 Model, Elevating Performance and Transitioning License Agreement to MIT.

Enhanced Efficiency: DeepSeek Unveils New V3 Model, Elevating Performance and Transitioning License Agreement to MIT.

DeepSeek has released a new version, DeepSeek-V3-0324, with a significant change of switching the license to MIT. The original DeepSeek model was already open-source but used its custom license. Now, it has transitioned to the widely open MIT License, allowing almost unrestricted usage.

In terms of model efficiency, DeepSeek-V3-0324 outperforms previous versions when tested on writing code in various languages, scoring a remarkable 55% accuracy in Aider’s Polyglot benchmark. Specifically, among models that don’t require deep thinking layers, it only trails behind Claude 3.7 Sonet with a 60.4% accuracy. Additionally, it comes with significantly lower running costs compared to others (1.12 dollars versus 17.72 dollars).

Source: Simon Willison

TLDR: DeepSeek introduces the enhanced DeepSeek-V3-0324 model with a switch to the more permissive MIT License, showcasing improved code writing capabilities and cost efficiency.

More Reading

Post navigation

Leave a Comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Phi-4 AI Unleashed by Microsoft: Audio Perception, Visual Recognition, Messaging Response Capabilities with Gemini 2.0 Flash Scoring.