💡 Leading inference providers — @basetenco, @DeepInfra, @FireworksAI_HQ, and @togethercompute — are cutting cost per token by up to 10x across industries with optimized inference stacks powered by the NVIDIA Blackwell platform. By combining #opensource frontier intelligence with NVIDIA Blackwell’s hardware–software codesign, and their own optimized inference stacks, these providers are delivering dramatic token cost reduction for businesses including @SullyAI, Latitude, Sentient, and Decagon. 🔗

⚡ Powered by NVIDIA Blackwell, @togethercompute and @DecagonAI are accelerating AI customer service — delivering human-like voice interactions in under 600 ms and cutting costs by 6x. With Together’s optimized inference stack running on NVIDIA Blackwell, Decagon powers real-time concierge experiences at scale — handling hundreds of queries per second with sub-second precision.

🩺 @SullyAI is transforming healthcare efficiency with Baseten’s Model API, running frontier open models like gpt-oss-120b on NVIDIA Blackwell GPUs. With their optimized inference stack built using NVIDIA Blackwell, NVFP4, TensorRT-LLM, and NVIDIA Dynamo, Baseten delivered a 10x cost reduction and 65% faster responses for key workflows such as clinical note generation.

@basetenco @DeepInfra @FireworksAI_HQ @togethercompute @DecagonAI @sullyai ⚙️ Latitude runs large-scale mixture-of-experts models on DeepInfra’s inference platform, powered by NVIDIA Blackwell GPUs, NVFP4, and TensorRT LLM. DeepInfra reduced cost per million tokens from $0.20 to $0.05 — a 4x efficiency gain.

To manage scale and complexity, @SentientAGI uses the Fireworks AI inference platform running on NVIDIA Blackwell. With @FireworksAI_HQ’s Blackwell-optimized inference stack, Sentient achieved 25-50% better cost efficiency compared with its previous Hopper-based deployment.  In other words, the company could serve 25–50% more concurrent users on each GPU for the same cost. The platform’s scalability supported a viral launch of 1.8 million waitlisted users in 24 hours and processed 5.6 million queries in a single week while delivering consistent low latency.