Hello, and welcome to this week’s changelog update! Let’s dive into what’s new:
-
60% Faster Time to Healthy and Deployment Promotion
We’ve enhanced our health check engine, reducing the time it takes to propagate the status of instances, deployments, and services by up to 60%. As a result, resource statuses are reported as soon as your application is ready, and deployment promotion is now significantly faster.
In our benchmark, we observed instances becoming healthy in around 10 seconds compared to 30 seconds previously.
-
Improved Traffic Shifting During Deployment Promotion
We’ve improved how traffic shifts during the promotion of a new deployment to deliver a better experience. Previously, traffic was briefly balanced between the current and upcoming deployments, which could lead to unexpected behavior. We now fully route to the new deployment once it’s ready, resulting in a more predictable and consistent experience.
-
Serverless Postgres Now Included in Usage Summary
You can now track your current billing cycle usage for the Serverless Postgres service from the Usage and Billing page in the control panel.
-
New One-Click Models: Magistral, DeepSeek-R1 0528 Qwen 3, Llama 4 Scout
Mistral just dropped their first reasoning model: Magistral — and it’s now available for one-click deployment on our serverless GPUs.
We’ve also added DeepSeek-R1 0528 Qwen3 8B and Llama 4 Scout to the one-click lineup. Deploy in one-click to get a dedicated GPU-powered endpoint for these state-of-the-art models.
