The Interconnects AI research institution has released its 2025 Open-Source Model Year in Review report, labeling it a milestone year for open-source model development. According to the report, open-source models have now matched the performance of closed-source models in most key benchmark tests. DeepSeek R1 and Qwen 3 were named the most influential models of the year. DeepSeek R1 was released on January 20th and is licensed under the MIT open-source license, showcasing the innovative capabilities of small teams and driving many Chinese laboratories to open their models. The Qwen 3 model family covers a wide range of areas, from general models to visual, programming, and embedded models, and has achieved outstanding performance in multilingual capabilities, with download volumes surpassing Llama to become the most popular fine-tuning base model. The report highlights the growing maturity of open-source models and their increasing adoption in various industries.