Case Study: Latent Health achieves 7× lower training costs and 97% clinical QA accuracy with Together AI Instant Clusters

Latent Health cuts training costs 7× and reaches 97% clinical QA accuracy with Together AI

Latent Health, an AI-native healthcare startup, needed to build high-accuracy clinical AI models for major health systems but faced unsustainable costs and infrastructure limitations. Their challenge was that AWS H100 instances were too expensive at $98/hour, and traditional cloud services prevented them from implementing complex, custom training workflows. They turned to Together AI's Instant GPU Clusters service for a solution.

By adopting Together AI's Instant Clusters, Latent Health gained direct SSH access to bare-metal H100s, enabling multi-node reinforcement learning and rapid training. This solution resulted in a 7x reduction in training costs ($14/hour), achieved 97% accuracy on clinical QA benchmarks—outperforming GPT-4—and sped up experimentation by 2-3x. This technical advancement directly translated to an 80% reduction in medication review times for their partner hospitals, significantly increasing pharmacist throughput and patient care efficiency.

Latent Health