Case Study: Vercept reached 5× performance vs. OpenAI and ~30% cost savings on Together AI
Key results
The challenge
Vercept builds AI that automates computer tasks, where small per-step latencies compound across long task chains and cause errors. Standard inference frameworks could not deliver the accuracy and price-performance the workload needed.
The solution
Vercept worked with Together AI to optimize and serve its models with autoscaling infrastructure and flexible term lengths, enabling a model-deployment cycle under 24 hours.
“There is no comparison on price—we easily save ~30% on costs for similar terms.”
LWLuca WeihsCo-founder, Vercept
The results, in context
Together AI's page reports Vercept achieved 92% accuracy on ScreenSpot v1 versus OpenAI's 18.3% (with similar gains on ScreenSpot v2 and GroundUI Web), delivering roughly 5× better performance than OpenAI on computer-automation tasks, alongside about 30% cost savings versus hyperscalers.