Faster Not Bigger: New R1T2 LLM Combines DeepSeek Versions

TNG Technology Consulting released DeepSeek-TNG R1T2 Chimera, a new large language model combining three previous DeepSeek releases. The model, designed for enterprise use prioritizing reasoning and concise answers, aims to reduce costs and preserve reasoning skills. R1T2, built using the “assembly of experts” approach, achieves 90-92% of R1-0528’s performance on reasoning benchmarks while generating shorter responses, reducing inference time by 60%.

*****
Written on