SiliconFlow

Score 8.0

An inference cloud for mainstream open models, popular with developers.

Built by: SiliconFlow · China startup
Categories: APIs & gateways
Pricing: Usage-based（Billed per token; some models have a free tier and new users get trial credit. See the official site.）
Open source: No
Self-hostable: No
International version: Yes
Data leaves China: Optional

"Data leaves China" means whether your data may be transferred outside mainland China when using the product.

Overview

SiliconFlow is a China-based inference cloud that turns mainstream open models — DeepSeek, Qwen, GLM, and more — into one unified API billed per token, with free tiers on some small models and trial credit for new users. Its value is "one API for many open models": developers can prototype, run batch jobs, and compare models without maintaining their own inference clusters or locking into one cloud vendor. The international edition (siliconflow.com) runs separately and serves overseas developers. Compared with Bailian or Volcengine Ark, SiliconFlow is lighter and focused on inference itself, with fewer enterprise add-ons (fine-tuning, private deployment, industry solutions); the team also contributes to open-source inference tooling, giving it credibility in the developer community. Good fit for individual developers, startups, and apps that call open models at volume.