[Tencent Hunyuan 3.0 Official Version Open-Sourced: Switched to Apache 2.0 to Remove Overseas Restrictions, Halved Hallucination Rate] According to a report by Jinse Finance, on July 6, Tencent officially released the 295B parameter Hunyuan 3.0 (Hy3) Mixture of Experts (MoE) model in its official version. The most significant change is the switch to the highly permissive Apache 2.0 open-source license, removing the previous regional restrictions that prohibited use in the EU, UK, and South Korea. When the preview version was released in April this year, strict regional restrictions and the 100 million monthly active commercial use threshold deterred many overseas or multinational teams. The official version has now cleared all compliance barriers and made hardcore optimizations for deployment challenges: it integrates a fast-and-slow thinking mechanism and incorporates a 3.8B parameter Multi-Token Prediction (MTP) layer for parallel generation, effectively reducing inference latency. Through fine-grained data cleaning and training constraints, the official version has reduced the hallucination rate of the large model from 12.5% to 5.4%, and the error rate in multi-turn interaction tests has dropped from 17.4% to 7.9%. To address the persistent issue of tool invocation errors during intelligent agent development, Hunyuan 3.0 has also undergone targeted enhancements. Whether in mainstream frameworks like Cline or CodeBuddy, cross-framework tool invocation accuracy fluctuations have been kept below 4%, resulting in more stable outputs. The officially released FP8 quantized version has also lowered the memory requirements for local deployment and fine-tuning.