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 Reflection's 1M-context Beam-1.0, its third new model API in eight days.

model-apisopen-modelslong-contextdeployment-safetyhosted-tools
◆Current state
Baseten's Model APIs now serve Reflection's Beam-1.0, with a 1M-token context window, tool calling and always-on reasoning tunable through reasoning_effort. It arrives a day after Meta Muse Spark Open and NVIDIA Nemotron 3.5 Super VL, and a week after DeepSeek V4.1 Flash Fast. Alongside the catalog growth came deployment controls: allow-listed egress via Truss config and rolling deployments on by default.
◆Where it's heading
Baseten is widening its hosted model catalog with new open and partner models within days of their release, while adding the operational controls enterprises require before running sensitive workloads. Catalog churn runs both ways: new models arrive and older ones are deprecated on a schedule. Hosted Tools web search remains the period's most directional move.
◆Prediction
Expect more new models in Model APIs at roughly weekly pace, some with Fast capacity variants, plus another deprecation round for older entries. The entries don't show pricing for the new models or which Hosted Tools type comes next.

◆Recent moves

  1. 8d ago

    Beam-1.0 available on Baseten

    Reflection's Beam-1.0 joins Model APIs with a 1M-token context, tool calling and adjustable reasoning effort. It continues the steady catalog additions and gives Baseten a long-context reasoning option alongside the open-weight models.

    View source ↗
  2. 9d ago

    Muse Spark Open and Nemotron 3.5 Super VL available on Baseten

    Two new open models, Meta's Muse Spark Open and the vision-language Nemotron 3.5 Super VL, join Model APIs on the OpenAI-compatible endpoint. It continues the pattern of fast catalog additions next to the deployment hardening.

    View source ↗
  3. 12d ago

    Egress restrictions

    Deployments can now restrict outbound traffic to an allow list of IP ranges and domains by setting egress_restrictions in Truss config. It's a concrete control for sensitive workloads and fits Baseten's run of security hardening, from b10_ key prefixes to OIDC for training jobs.

    View source ↗
  4. 12d ago

    Rolling deployments enabled by default

    New model environments now default to rolling deployments, replacing replicas incrementally and shifting traffic in steps; existing environments keep their settings. Making the safer promotion path the default lowers outage risk on model updates, part of the same operational-maturity push as the stable CLI.

    View source ↗
  5. 16d ago

    DeepSeek V4.1 Flash Fast available on Baseten

    DeepSeek V4.1 Flash is now also offered on dedicated Fast capacity, serving the same 552B multimodal model at full precision. It turns a catalog addition into a tiered offering, letting latency-sensitive agentic workloads pay for headroom.

    View source ↗
  6. 22d ago

    Baseten CLI 1.1 adds model rename from the terminal

    CLI 1.1 adds baseten model rename, which changes a model's name without touching its ID, endpoints or deployments. A small but practical addition two weeks after the CLI was declared stable.

    View source ↗