Models

Serenity Edge serves two model families over the same deployments. Every model is served in NVFP4 precision with a context window of 131,072 tokens and a maximum output of 32,768 tokens.

Catalogue

Model id	Family	Architecture	Context	Max output	Best for
orion-pro	Orion	Dense, 27B parameters	131,072	32,768	Quality-sensitive work: long documents, complex instructions, tool calling
orion-plus	Orion	Mixture of experts, 35B total / 3B active	131,072	32,768	High-volume, latency-sensitive work: chat, classification, extraction
qwen3.8-27b	Open	Dense, 27B parameters	131,072	32,768	Same deployment as orion-pro, addressed by its open model name
qwen3.6-35b-a3b	Open	Mixture of experts, 35B total / 3B active	131,072	32,768	Same deployment as orion-plus, addressed by its open model name

Pricing is listed on Pricing and limits. Use GET /v1/models to retrieve the catalogue programmatically; see List models.

Two families, one set of deployments

Open models

qwen3.8-27b and qwen3.6-35b-a3b are the open Qwen models, served as published. Use these ids when you want to name the underlying open model explicitly, for example to keep parity with a deployment elsewhere or to compare results with other providers.

Orion

orion-pro and orion-plus are the Orion family. Today they resolve to the same deployments as the open models above, wrapped in the Serenity serving recipe: the sampling defaults, prompt handling and serving configuration that Substrate AI tunes for its workloads. Address the Orion ids when you want to follow Serenity's recommended configuration and benefit from future improvements without changing your code.

Preview

Substrate AI is preparing its own fine-tuned Orion models. When they ship, the orion-pro and orion-plus ids will resolve to them. The change will be announced in the release notes in advance, together with the version of the open models that stays available under the qwen ids.

Choosing between pro and plus

• Start with orion-plus (or qwen3.6-35b-a3b). It activates only 3B parameters per token, which makes it fast and inexpensive, and it handles most conversational, classification and extraction workloads well.
• Move to orion-pro (or qwen3.8-27b) when the task needs more depth: long multi-document reasoning, strict adherence to complex instructions, or reliable multi-step tool calling.

Both models support streaming, tool calling, structured outputs and reasoning. See the API reference for the full parameter list.

Precision

All models run in NVFP4, a 4-bit floating-point format for NVIDIA GPUs. It is the standard serving precision on Serenity Edge and is what the published prices assume.