Fast rollouts for models
that aren’t finished yet.
- A drafter trained online, on your own rollouts, as they run.
- Merged or unmerged adapters at frontier MoE scale, no rank caps
- Weight sync to a live endpoint in seconds
- Token-identical to unaccelerated inference
Research preview & design partners
Blink is in research preview. We’re working closely with a small group of design partners running offline RL, online RL, and large eval sweeps.
Bring your models, harness, and real workloads. We’ll work with your team to shape the rollout infrastructure around what you need.
Blink is in research preview with a small group of design partners. Bring your RL and eval workloads; help shape what we build.
hello@runblink.ai