TNG: DeepSeek R1T2 Chimera

DeepSeek-TNG-R1T2-Chimera is the second-generation Chimera model by TNG Tech. This 671-billion-parameter Mixture-of-Experts (MoE) text-generation model is built from DeepSeek-AI’s R1-0528, R1, and V3-0324 checkpoints using an Assembly-of-Experts merge. This unique tri-parent design delivers exceptional reasoning performance while operating significantly faster than its predecessors, making it highly efficient. The model runs approximately 20% faster than the original R1 and over 2x faster than R1-0528 under vLLM, providing an excellent cost-to-intelligence ratio. It supports context windows up to 60,000 tokens in standard use (tested up to ~130,000 tokens) and maintains consistent <think> token behavior. With pricing at $0.25/0.85 per 1M tokens (input/output) and a max output of 4K tokens, it's perfect for long-context analysis, complex dialogue, and other open-ended generation tasks. Access this PRO-tier model on Multi AI today!