Supported Models

Xinference Cloud offers models in two forms: Hosted models that run serverlessly on Xinference infrastructure and are billed per token, and Dedicated models that run on GPU capacity reserved for your organization.

The console is the authoritative catalog

The model catalog changes frequently — models are added, promoted, and retired between releases. The lists on this page are a snapshot taken on 11 August 2026, not a live inventory. Always confirm current availability and pricing in the Xinference Cloud console, or via GET /api/v1/deployable-models.

Hosted Models

Serverless models running on Xinference infrastructure, priced per million tokens. No deployment is required — a hosted model marked Ready can be used immediately.

Available in the console as of 11 August 2026:

Model Identifier Status Notes
Qwen3.5 0.8B qwen · qwen3.5 Ready Chat and general assistant workloads. Input $1.00 / M tokens, output $2.00 / M tokens.
DeepSeek V4 Flash deepseek · deepseek-v4-flash Contact support Low-latency chat variant.
GLM-5.1 zhipu · glm-5.1 Contact support Chat, reasoning, coding, and tool-use workloads.

A status of Contact support means the model is listed but not self-serve — reach out to have it enabled for your organization.

Dedicated Models

Launch a model on your own reserved GPU infrastructure for isolated, dedicated capacity. Dedicated models are deployed and billed by running time rather than per token — see Create a Deployment →.

The console groups dedicated models into these categories:

  • Language
  • Embedding
  • Search Ranking
  • Image
  • Audio
  • Video
  • Custom

Featured in the console as of 11 August 2026:

Model Family Provider Capabilities
Gemma 4 Gemma Google Chat, vision, multimodal — instruction model for multimodal chat and assistant workloads
Qwen3.5 0.8B Qwen3.5 Qwen Chat, instruct — general assistant workloads

Both are marked Optimized, indicating tuned defaults for the selected instance type.

Model families in the catalog

Families present in the Xinference Cloud catalog. Not every variant is enabled for every organization — check the console for what you can launch:

Family Variants Type
qwen3.5 0.8B, 2B, 4B, 9B, 27B, 35B, 122B, 397B Language
qwen3-vl Vision-language
gemma-4 Multimodal
glm 4.5, 4.7, 5.1 Language
deepseek-v4 Standard, Flash Language
bge-m3 Embedding

GPU instance types

Dedicated deployments run on AWS GPU instances. Xinference selects a recommended instance for each model and variant; the instance families in use are:

Family Sizes in use GPU
g5 xlarge, 2xlarge NVIDIA A10G
g4dn 2xlarge, 4xlarge, 8xlarge, 12xlarge, 16xlarge NVIDIA T4

The recommended instance for a given model is set in the catalog and can change between releases — the console shows the current recommendation when you launch. See Deployment Overview → for how instance selection affects billing.