Laya

Laya is an open-source decision model published under Apache 2.0 — a direct answer to closed decision APIs. Instead of calling a vendor endpoint, you run the model yourself and keep both the input context and the decisions on your own infrastructure.

What it is

Laya is an encoder-based model for constrained decisions rather than free-form generation. Its public repository and model card describe an openly licensed model that evaluates defined decision questions in a single forward pass.^[1][2]

What it returns

The documented interface covers choice, score, and noul questions. It returns typed decision values and probability information for those questions rather than generated prose.^[2]

Self-hosting

Laya is released under the Apache License 2.0 and published as model weights you can run on infrastructure you control.^[1]Self-hosting keeps the request context and resulting decisions inside your deployment boundary, while making your team responsible for serving, operations, and evaluation.

How it compares

OpenAI's Decisions API is a hosted OpenAI endpoint, while TypeSafe Jev is a hosted, purpose-built decision model. Laya occupies the self-hosted position: its open license and published weights favor infrastructure control over vendor-managed operation. See the full decisions api vs jev comparison for the product-by-product trade-offs.

What's not confirmed

The available first-party material does not establish every field needed for a like-for-like product comparison. We leave those fields open instead of inferring them from other decision models.

• Pricing: Not yet confirmed.
• Input modality: Not yet confirmed.

Latency is not included in that list: the vendor model card reports about 39.5 ms per question on a T4 GPU. Treat it as a vendor measurement rather than an independent benchmark.^[2]