API Key Setup10 min readPublished: 2026-07-28Updated: 2026-08-21

SiliconFlow API Key Complete Guide (2026): Setup, Free Models, Pricing & BYOK

Q: Is SiliconFlow API free?

SiliconFlow offers both free and paid models. Models marked free on the live pricing page are billed at zero within fixed rate limits. The current official announcement states a ¥16 China-site or $1 global-site new-user credit.

Q: What is the SiliconFlow API base URL?

The Chinese endpoint is https://api.siliconflow.cn/v1 and the global endpoint is https://api.siliconflow.com/v1. Both accept the same API keys and model IDs. Use the .com domain for better latency outside China.

Q: Is SiliconFlow compatible with the OpenAI SDK?

Yes, fully compatible. Use the standard OpenAI SDK (Python, JavaScript, Go, etc.), set the base URL to the SiliconFlow endpoint, and use your SiliconFlow API key. The request and response format is identical to OpenAI's — the only difference is the model ID format (vendor/model-name).

Q: How many models does SiliconFlow offer?

Over 200 open-source models from providers including DeepSeek, Alibaba Qwen, Zhipu GLM, Moonshot Kimi, and MiniMax. The catalog is continuously updated as new models are released. You can browse the full list on the SiliconFlow console's model page.

Q: What is BYOK and how does it work with SiliconFlow?

BYOK (Bring Your Own Key) means you paste your SiliconFlow API key into third-party AI tools like Cherry Studio, ChatHub, or immersive-translate, and use SiliconFlow's models as the backend. Because the endpoint is OpenAI-compatible, most tools that support custom API endpoints work out of the box.

Everything about SiliconFlow API keys: register, use current free models, understand pricing and fixed free-model limits, call the live model catalog through an OpenAI-compatible endpoint, and set up BYOK.

By Heizi· Founder & Editor· Published: 2026-07-28· Updated: 2026-08-21

What is SiliconFlow?

SiliconFlow (硅基流动) is a model inference platform that exposes a broad, changing catalog through one API key and an OpenAI-compatible endpoint.

The platform handles model hosting, load balancing, and inference optimization, so you don't need to deploy models on your own GPUs. This makes SiliconFlow especially popular among Chinese developers and anyone who wants to experiment with multiple open-source models without managing infrastructure.

SiliconFlow offers a generous free tier: selected models are billed at zero inside fixed rate limits after identity verification. The 2026 free-tier guide explains how that works, what is still free, and when to switch to paid.

Available models: use the live catalog

SiliconFlow aggregates models from dozens of leading open-source providers. The catalog includes the DeepSeek family (DeepSeek V4-Pro, DeepSeek-R1 reasoning models), Alibaba's Qwen3 series, Zhipu's GLM-5, Moonshot's Kimi K2.6, and MiniMax-M2.5. Each model is identified by a namespaced ID like deepseek-ai/DeepSeek-V4-Pro or Qwen/Qwen3-8B.

A standout feature is the free model tier. Qwen3-8B is ideal for general conversation and light tasks, DeepSeek-R1-Distill-Qwen-7B provides chain-of-thought reasoning at no cost, and DeepSeek-OCR handles document and image text extraction for free. These are not toy models — they are production-capable and only limited by rate limits.

For a dated list of model IDs currently marked free, see the SiliconFlow free models snapshot. How the free tier itself works — verification, fixed limits, and credit vs free models — is in the SiliconFlow API free tier 2026 guide.

How to Get a SiliconFlow API Key

Registering and obtaining an API key takes about five minutes. Visit siliconflow.cn, sign up with a phone number (Chinese +86 numbers are standard; international users can use the global site at siliconflow.com), complete real-name verification to unlock bonus credits, then navigate to the API Keys page in the console to create your key.

SiliconFlow's official June 2026 announcement states that new users can receive ¥16 on the China site or $1 on the global site. Identity verification is required to use all free models.

For the complete step-by-step walkthrough including screenshots of every page, the real-name verification process, and how to claim bonus credits, follow our SiliconFlow API key registration tutorial.

Pricing and Rate-Limit Tiers

SiliconFlow uses a prepaid credits model with a six-tier rate-limit system. Your tier is determined by your cumulative spending: L0 (under ¥50) has the lowest limits, while L5 (¥10,000+) unlocks the highest throughput. Moving up tiers is automatic — spend more, get higher limits.

Paid model prices change independently. Use SiliconFlow's live pricing page for every production estimate. Free models are billed at zero, but their rate limits are fixed rather than scaling with paid usage tiers.

The complete pricing breakdown compares every model side by side so you can choose the most cost-effective option for your workload.

Free models: Qwen3-8B, DeepSeek-R1-Distill-Qwen-7B, DeepSeek-OCR (rate-limited)
Paid models: verify input, output and cache prices on the live pricing page
Six rate-limit tiers: L0 (<¥50 spend) to L5 (≥¥10,000 spend)
Current new-user offer: ¥16 on the China site or $1 on the global site
Prepaid credits model — top up as needed, no monthly subscription required

Quick Start: Your First SiliconFlow API Call

SiliconFlow is fully OpenAI-compatible. Use the standard OpenAI SDK with the SiliconFlow base URL and your API key. The model ID format is vendor/model-name.

python

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-siliconflow-key",
    base_url="https://api.siliconflow.cn/v1"
)

response = client.chat.completions.create(
    model="Qwen/Qwen3-8B",  # free model
    messages=[
        {"role": "user", "content": "用一句话解释什么是 API"}
    ]
)

print(response.choices[0].message.content)

Calling the free Qwen3-8B model through SiliconFlow using the OpenAI Python SDK

javascript

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.SILICONFLOW_API_KEY,
  baseURL: "https://api.siliconflow.cn/v1",
});

const response = await client.chat.completions.create({
  model: "deepseek-ai/DeepSeek-V4-Pro",
  messages: [{ role: "user", content: "Write a haiku about APIs" }],
});

console.log(response.choices[0].message.content);

Calling the DeepSeek-V4-Pro model with the OpenAI JavaScript SDK

Using SiliconFlow in AI Tools (BYOK Mode)

One useful feature is Bring-Your-Own-Key (BYOK) compatibility. Tools that accept a custom OpenAI-compatible endpoint can use your SiliconFlow key and the models currently listed in the platform catalog.

Popular BYOK targets include Cherry Studio (multi-model chat client), ChatHub (browser extension), immersive-translate (document and webpage translation), and Claude Code (using SiliconFlow as the backend for coding agents). In each tool, you simply set the base URL to https://api.siliconflow.cn/v1, paste your key, and pick a model.

Our SiliconFlow BYOK setup guide walks through the exact configuration for each major tool with screenshots.

Common Pitfalls and How to Avoid Them

The most common error is using the wrong model ID format. SiliconFlow requires the full namespaced ID (vendor/model-name), not just the model name. For example, Qwen3-8B alone will fail — you must use Qwen/Qwen3-8B.

Second, international users sometimes confuse the Chinese endpoint (api.siliconflow.cn) with the global one (api.siliconflow.com). They accept the same keys but route through different servers. Use the .com domain if you are outside China for better latency.

Third, free models have stricter rate limits than paid ones. If you hit 429 errors on a free model during testing, either wait for the limit to reset or switch to a low-cost paid model like Qwen3-235B for heavier workloads.

Best Practices

Start with the free models to prototype your application at zero cost. Once you validate the use case, move to paid flagship models like DeepSeek-V4-Pro for production. This two-stage approach minimizes spending during development.

Store your API key as a server-side environment variable and never commit it to version control. Use descriptive key names in the SiliconFlow console so you can track which project each key belongs to.

Monitor your credit balance through the console dashboard and set up recharge reminders. For production applications, implement retry logic with exponential backoff for 429 rate-limit responses, and consider upgrading to a higher spending tier for more headroom.

FAQ

Is SiliconFlow API free?

SiliconFlow offers both free and paid models. Models marked free on the live pricing page are billed at zero within fixed rate limits. The current official announcement states a ¥16 China-site or $1 global-site new-user credit.

What is the SiliconFlow API base URL?

The Chinese endpoint is https://api.siliconflow.cn/v1 and the global endpoint is https://api.siliconflow.com/v1. Both accept the same API keys and model IDs. Use the .com domain for better latency outside China.

Is SiliconFlow compatible with the OpenAI SDK?

Yes, fully compatible. Use the standard OpenAI SDK (Python, JavaScript, Go, etc.), set the base URL to the SiliconFlow endpoint, and use your SiliconFlow API key. The request and response format is identical to OpenAI's — the only difference is the model ID format (vendor/model-name).

How many models does SiliconFlow offer?

Over 200 open-source models from providers including DeepSeek, Alibaba Qwen, Zhipu GLM, Moonshot Kimi, and MiniMax. The catalog is continuously updated as new models are released. You can browse the full list on the SiliconFlow console's model page.

What is BYOK and how does it work with SiliconFlow?

BYOK (Bring Your Own Key) means you paste your SiliconFlow API key into third-party AI tools like Cherry Studio, ChatHub, or immersive-translate, and use SiliconFlow's models as the backend. Because the endpoint is OpenAI-compatible, most tools that support custom API endpoints work out of the box.