TNG Unveils “Wild” Version of DeepSeek R1T2: Outpaces Official Release, Open Source

The German AI company TNG has launched the “DeepSeek R1T2” model. Developed based on three foundational models—DeepSeek R1, R1-0528, and V3—it employs the Mixture of Experts (MoE) technology, achieving a 20% speed improvement over R1 while significantly enhancing reasoning capabilities. Compared to the original model, R1T2 demonstrates a substantial boost in intelligence, resolving token consistency issues and delivering a natural conversational experience even without system prompts.