One playground for models and compute
The Preview Playground now has its own console navigation entry. Switch between Chat, Decisions, Audio, Embeddings, Code, Agents and Sandboxes. Code and sandbox launches retain the resource review; agents retain their saved model and budget limits.
Models and Playground share a persistent view switch. Open a model to try it, then return to the same filtered grid without losing your draft or selected model. The grid keeps publisher colors, logos and pricing, with Vision, Tools, Reasoning, Decisions, Audio and Embeddings filters. Automatic routing is available in a compact row above the cards.
The decision view has a text and JSON input editor, editable yes/no, choice and rating questions, and side-by-side answers with probabilities, latency and receipt costs. Compare up to three models or explicitly repeat a comparison ten times. Each request is billed separately; stopping prevents unsent requests, while requests already sent may still finish and be billed.
Span-02, Kev 4B, Blink v0.3, d3, Decider v1.1, REAX 1 and REAX 1.2 expand the eligible open-weight catalog to eighteen decision models. Serving remains gated on serving readiness and live qualification. The console shows each available model’s current input price; decision requests have no output-token charge.
Explore Playground or read the decision guide.
Conversational inference now includes a shared Nodus Compute service-identity instruction across SDK, endpoint and playground calls. It preserves model attribution and customer instructions, and its input is included in request limits and usage.