Together AI Achieves Breakthrough Inference Speed with NVIDIA’s Blackwell GPUs
Together AI has unveiled the world's fastest inference performance for the DeepSeek-R1-0528 model, leveraging NVIDIA's HGX B200 platform. This advancement solidifies its position as a leading platform for scalable open-source reasoning models.
Early testing with select clients, including Zoom and Salesforce, demonstrated significant performance gains. The full rollout of Nvidia Blackwell GPU support now enables unprecedented serverless inference speeds, as confirmed by benchmarks dated July 17, 2025.
The technological breakthrough stems from a fully optimized stack featuring custom GPU kernels and proprietary inference architecture. State-of-the-art speculative decoding and model optimization techniques deliver enhanced efficiency without quality compromise.
