Kimi K3
V3 Ultra-Fast Version,The current price is a limited-time 50% discount and will return to the original price on July 31st. The original price is: input: $0.55/M, output: $2.2/M. The model provider is the Sophnet platform. DeepSeek V3 Fast is a high-TPS, ultra-fast version of DeepSeek V3 0324, featuring full-precision (non-quantized) performance, enhanced code and math capabilities, and faster responses! DeepSeek V3 0324 is a powerful Mixture-of-Experts (MoE) model with a total parameter count of 671B, activating 37B parameters per token. It adopts Multi-Head Latent Attention (MLA) and the DeepSeekMoE architecture to achieve efficient inference and economical training costs. It innovatively implements a load balancing strategy without auxiliary loss and sets multi-token prediction training targets to enhance performance. The model is pre-trained on 14.8 trillion diverse, high-quality tokens and further optimized through supervised fine-tuning and reinforcement learning stages to fully realize its capabilities. Comprehensive evaluations show that DeepSeek V3 outperforms other open-source models and rivals leading closed-source models in performance. The entire training process only requires 2.788M H800 GPU hours and remains highly stable, with no irrecoverable loss spikes or rollbacks.