SiliconFlow logo

SiliconFlow

★★★★★4.2PlatformCNPaidUpdated 27 Oct 2026

Inference API for open models, priced in CNY per token

SiliconFlow hosts open-weight models — DeepSeek, Qwen, GLM, image and video models — behind one OpenAI-compatible API.

siliconflow.cn
SiliconFlow homepage screenshot

Starter prompts

Copy one into SiliconFlow and swap the {braces} for your own details.

  • Compare {model A} and {model B} for {use case} on cost, latency and output quality.
  • Estimate what it costs to run {task} at {n} requests per day on this platform.
  • Sketch the API call needed to {task}, including auth and error handling.

Overview

SiliconFlow (硅基流动) is the Chinese equivalent of a hosted inference marketplace: one key, one OpenAI-compatible endpoint, and a catalogue of open-weight models you would otherwise have to deploy yourself. Several small models are free to call, which makes it a cheap way to prototype without a GPU.

Access & Availability

Signing up
Mainland mobile number required for the .cn console.
Availability
API endpoints respond worldwide; the console is mainland-oriented.
Language
Chinese console; API docs have an English version.
Billing
CNY via Alipay; invoicing available for companies.

Key Features

OpenAI-compatible API

Point an existing SDK at their base URL and change the model name.

Wide catalogue

Text, embedding, reranking, image and video models in one place.

Free tiers on small models

Some models are callable without payment, with rate limits.

How to Use

  1. Register with a mainland mobile number.
  2. Create an API key in the console.
  3. Swap the base URL in your OpenAI client and pick a model from the catalogue.

Core Advantages

  • ✓No infrastructure to run
  • ✓Pricing is transparent and per token
  • ✓Good coverage of Chinese open-weight models

Who It's For

Developers prototyping with open modelsTeams avoiding GPU procurementAnyone benchmarking DeepSeek or Qwen variants

Pricing & Free Trial

Usage-based Read from the vendor page
PlanPriceNotes
Free modelsFreeRate-limited, small models only
Pay as you goCNY per million tokensPer-model pricing

Free tier: A signup credit grant plus free calls on selected small models with rate limits.

Last checked 28 Oct 2026 · source page

Prices are in CNY and differ by model; the catalogue page is the authoritative list.

Frequently asked questions

What is SiliconFlow?
SiliconFlow hosts open-weight models — DeepSeek, Qwen, GLM, image and video models — behind one OpenAI-compatible API.
Is there a free plan for SiliconFlow?
A signup credit grant plus free calls on selected small models with rate limits.
How much does SiliconFlow cost?
SiliconFlow uses a usage-based model — see the pricing section for the plans we have on file.
What can SiliconFlow do?
Its headline capabilities: openai-compatible api; wide catalogue; free tiers on small models.
Who is SiliconFlow for?
It is aimed at developers prototyping with open models, teams avoiding gpu procurement, anyone benchmarking deepseek or qwen variants.
Is SiliconFlow available in mainland China?
Yes — SiliconFlow is built in mainland China and is designed to work there directly. See the access notes above for signup and billing details.

Our Verdict

The least painful way to call a Chinese open model from outside China — one endpoint, no procurement. Check the terms before sending sensitive data.

Community rating

4.7
14 ratings

Sign in to share your rating with the community.

More like this

Compare SiliconFlow with alternatives

Put SiliconFlow next to its closest rivals — ratings, free tiers and entry prices in one table:

  • OpenRouter logoOpenRouter★ 4.6
  • Hugging Face logoHugging Face★ 4.6
  • Google AI Studio logoGoogle AI Studio★ 4.5
Open the comparison table