LoginSign Up
← Back to all sparks
B

Baseten

AI-ASSISTANTS
Velocity5.0

AI model deployment and inference platform for running ML models in production.

www.baseten.co ↗

Baseten adds AWS AssumeRole auth and a Viewer role — enterprise governance for model inference.

model-inferenceenterprise-authaws-integrationautoscalingglm
◆Current state
Baseten is tightening its enterprise access layer while expanding the model catalog. The new Viewer role adds read-only access for teammates who need to invoke models and inspect configurations without deployment permissions. AWS AssumeRole authentication eliminates long-lived credentials for pulling private base images from ECR or model weights from S3. On the model side, three GLM 5.3 variants from Z.ai arrived within two days, and autoscaling schedules now let teams pre-warm capacity before traffic arrives.
◆Where it's heading
Baseten is positioning as the enterprise-grade inference platform for teams running production AI workloads with AWS-native infrastructure. The IAM-style access control additions (Viewer role, AssumeRole) are more characteristic of production deployments than dev/test usage. The autoscaling schedule feature suggests a customer base with predictable traffic patterns — think inference APIs, not exploratory experiments.
◆Prediction
AWS AssumeRole will likely expand to GCP and Azure IAM next, making cross-cloud model serving a differentiator for enterprise teams that already run multi-cloud workloads.

◆Recent moves

  1. 8d ago

    Viewer role for read-only access

    A new Viewer role lets teammates access Baseten resources — invoke models, inspect configurations — without deployment or write permissions. This fills the gap between full access and no access, enabling wider team visibility without security risk.

    View source ↗
  2. 8d ago

    AWS AssumeRole authentication

    Baseten can now assume an IAM role in a customer's AWS account for ECR image pulls and S3 weight mirroring during builds, replacing the need for long-lived access keys. This is the AWS-native security pattern enterprise teams require before allowing a third party to touch their private model artifacts.

    View source ↗
  3. 12d ago

    GLM 5.3 available on Baseten

    GLM 5.3, Z.ai's flagship model with a 1M-token context window, is now available through Baseten's OpenAI-compatible endpoint. The reasoning effort control (low/high/max) gives developers cost-vs-quality flexibility from the same model.

    View source ↗
  4. 14d ago

    GLM 5.3 Flash available on Baseten

    GLM 5.3 Flash is available via the OpenAI-compatible endpoint — a faster, lower-cost variant of Z.ai's latest model. Expanding the GLM model family gives Baseten users a full performance tier to choose from within the same model series.

    View source ↗
  5. 14d ago

    GLM 5.3 Fast: third tier of the GLM 5.3 family

    GLM 5.3 Fast (a third speed/cost tier in the GLM 5.3 family) is available on the same OpenAI-compatible endpoint. Rounds out the GLM 5.3 lineup already added this week.

    View source ↗
  6. 15d ago

    Autoscaling schedules

    Autoscaling schedules are now generally available, letting environments automatically adjust capacity on a defined time window. Pre-warming capacity before predictable traffic spikes reduces cold-start latency without manual intervention before each high-demand period.

    View source ↗