The use case: how Sword Health deploys Nebius AI Cloud
Sword Health runs its Dawn AI wellbeing specialist on Nebius Token Factory dedicated NVIDIA Blackwell endpoints, using custom speculative decoding to serve a much larger reasoning model alongside its MindGuard guardrail model. It also migrated its AI Vision models to Nebius cloud for its musculoskeletal care product.
More reported results
Figures published in Nebius AI Cloud’s case study. They are not independently measured by usedby.
Achieving a draft model acceptance rate of over 50% meant Dawn could make the leap to a model with over 200B parameters, roughly eight times larger, without becoming unbearably slow in production.
In the customer’s words
Designing Dawn required access to specialized infrastructure, health-specific governance, genuine clinical expertise, and a lot of specialized data. It requires clinical-grade reasoning from a highly capable model, but also system stability, reliability and consistency across every single interaction. We want every conversation to build long-term trust.Nebius AI Cloud users in HealthTech & Digital Health
Companies in the same sector, with public evidence.
See all Nebius AI Cloud customersGo straight to the source
Based on Nebius AI Cloud’s published case study. We check the public evidence; the results are reported by the vendor.
nebius.com- Last checked
- Nov 9, 2026
Open source capture What the story claims, and what we checked
What we compared with the page.
- The figure: under 12Checked
Printed word for word on the page, near the name of Sword Health.
- The passage quoted aboveChecked
Copied word for word from the page, near the name of Sword Health.
- Sword Health uses Nebius AI CloudChecked
Confirmed line.
- The result itselfNot checked
We quote it; we did not measure it.
The Nebius Token Factory team deployed custom speculative decoding trained specifically for Sword Health on dedicated NVIDIA Blackwell endpoints. Achieving a draft model acceptance rate of over 50% meant Dawn could make the leap to a model with over 200B parameters, roughly eight times larger


