Hosted in Australia
Startups and digital natives
Frontier open models for startups, priced below big cloud.
The Model API puts open models behind one OpenAI-compatible API. Switch in two lines, then grow from tokens to dedicated GPUs on the same API.
The margin problem
Your inference bill is your gross margin.
Every request you serve is a cost of goods sold. We price the same open models below Bedrock.
Output, per 1M tokens
DeepSeek V3.2Input, per 1M tokens
DeepSeek V3.2AWS Bedrock list prices, Asia Pacific (Sydney) region, Standard tier, USD per 1M tokens, 28 September 2026. Xinference vs Bedrock.
Developer experience
Switch in two lines. Keep your code.
- Chat completions and streaming
- Function calling, JSON mode and streaming
- TTFT and TPOT measured on every request
client = OpenAI(
- base_url="https://api.openai.com/v1",- api_key=OPENAI_API_KEY,+ base_url="https://api.xinference.co/v1",+ api_key=XINFERENCE_API_KEY,)Use cases
What startups build on Xinference.
Agents and coding tools
Open models with function calling and structured output.
RAG and search
Answer questions over your own documents with long-context models.
Voice agents
Real-time transcription and speech, on Dedicated Inference.
High-volume pipelines
Classification and extraction, where the per-token price is your margin.
Sovereignty as a sales asset
Win enterprise deals with a clear data answer.
Selling to banks, insurers, hospitals or government? Their first question is where the data goes.
Your environment, when needed
Move a workload to dedicated GPUs or your own cloud account, on the same API.
No training on your data
No prompt or response content kept by default.
No lock-in
Open weights behind an API you already know.
Scale path
From first token to dedicated GPUs, on the same API.
- 01
Model API
Pay per token. First request in minutes.
- 02
Dedicated Inference
Dedicated GPU capacity for steady traffic.
- 03
Your infrastructure
Your cloud or data centre, when a customer contract needs it.
Startup questions
Which inference providers are priced below AWS Bedrock?
Xinference serves the same open models below Bedrock’s list prices. Read the Bedrock comparison.
Can we use one SDK for chat completions, embeddings and images?
Chat completions, function calling and streaming run through the OpenAI SDK you already use. Image and speech models run on Dedicated Inference behind the same API.
How is the Model API billed?
Per token, per model, in USD per 1M tokens. Kimi K3 is $3.00 in and $15.00 out. DeepSeek V4.1 Flash is $0.30 in and $1.20 out. See pricing.
How does token pricing compare to dedicated GPUs?
Per token suits early or uneven traffic. At steady volume you can move to a set cost per GPU hour. The move is a base URL change, and an engineer can model both on your usage.
Which models can we use?
The Model API serves Kimi K3 and DeepSeek V4.1 Flash. Dedicated Inference can run other open models, plus your own fine-tunes. Don’t see the model you need? Talk to us about adding it.
Is our data used for training?
No. No training on your data, and no prompt or response content kept by default.
What are the rate limits?
Default limits apply per model. Higher concurrency is available on request through sales.
Last updated