LoginSign Up
← Back to all sparks
B

Baseten

AI-ASSISTANTS
Velocity6.3

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

www.baseten.co ↗

Baseten CLI 1.0.0 ships a stable command contract as regional deployments unlock enterprise compliance use cases.

ml-inferenceenterprise-compliancecli-stabilitymodel-servingdata-residency
◆Current state
Baseten is assembling the operational scaffolding enterprise buyers require before committing to a model-serving vendor: a stable CLI (v1.0.0) that ends breaking changes in the primary toolchain, regional deployment controls for data residency compliance, OIDC/AWS AssumeRole for credential-free training jobs, and a billing API for per-day model spend. Concurrent model churn—deprecating GLM and Kimi variants while adding DeepSeek V4.1 Flash—reflects active catalog management as the hosted-inference market shifts.
◆Where it's heading
Baseten is moving from ML infrastructure for ML teams toward a compliant model-serving platform for enterprise buyers. The combination of stable interfaces, regional isolation, and granular IAM authentication addresses the checklist items that security and compliance reviewers require. The next logical additions are audit logging and SSO/SAML.
◆Prediction
Baseten will add audit trail and SSO integration in the next quarter, completing the enterprise procurement checklist for regulated industry buyers.

◆Recent moves

  1. 17h ago

    Baseten CLI 1.0.0

    ⚡ SPARK

    The Baseten CLI reaches 1.0.0 with a stable command surface, marking the end of breaking changes in the primary interface for model deployment automation. This is a contract commitment, not a feature release—the signal is that the CLI is now safe to embed in CI/CD pipelines without tracking upstream changes.

    View source ↗
  2. 3d ago

    Baseten deprecates GLM 4.7, Kimi K2, Inkling, and DeepSeek v4 Pro on September 25

    GLM 4.7, Kimi K2.7/K2.6, Inkling, Inkling Small, and DeepSeek v4 Pro are being deprecated on September 25th. Removing underutilized model variants keeps the API catalog manageable and is consistent with Baseten's pattern of rotating models as successor versions land.

    View source ↗
  3. 4d ago

    OIDC and AWS AssumeRole for training jobs

    Training jobs can now authenticate via OIDC or AWS AssumeRole to pull private container images and download weights without storing long-lived credentials in Baseten. This removes a security anti-pattern that previously required teams to embed static cloud credentials to access private assets during builds.

    View source ↗
  4. 4d ago

    Model API costs

    A new Management API endpoint returns per-day model API spend subtotals, giving teams programmatic access to billing data without manual dashboard exports. For organizations with chargeback requirements or cost monitoring pipelines, this fills a gap that previously required manual reconciliation.

    View source ↗
  5. 5d ago

    DeepSeek V4.1 Flash available on Baseten

    DeepSeek V4.1 Flash, a 552B-parameter multimodal model built for agentic and coding workflows, is now available via Baseten Model APIs. Model catalog additions like this follow Baseten's pattern of adding frontier models shortly after release, giving inference customers immediate access without self-hosting.

    View source ↗
  6. 5d ago

    Regional deployments

    Regional deployments allow every replica in a deployment to be pinned to a specific Baseten region, supporting data residency and compliance requirements. Configurable via the dashboard, CLI, or Management API, this also addresses latency-sensitive workloads that benefit from geographic proximity.

    View source ↗