SiliconFlow 硅基流动
One OpenAI-compatible API for 170+ models, billed per token
Overview
SiliconFlow (硅基流动) is a Beijing inference platform that serves more than 170 language, image, video, speech and embedding models from DeepSeek, Qwen, GLM, Kimi and others through one OpenAI-compatible, pay-per-token API. It runs siliconflow.cn for China and siliconflow.com from Singapore, and filed for a Hong Kong listing in June 2026.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- 1,049K context
- Pay as you go, postpaid
- OpenAI-compatible API
- 1,049K context
- Serverless, pay per use
- 1,049K context
- 262K max output
- 1,049K context
- Serverless, pay per use
- Pay per image
- Guaranteed GPU capacity for stable performance
- Predictable billing
What you can produce with SiliconFlow 硅基流动
- One OpenAI-compatible API across 170+ language, image, video, speech and embedding models
- Per-token or per-image billing with no subscription, postpaid on the international site
- Free access to small models such as Qwen2.5-7B-Instruct and BAAI embedding models
- Rate limits that rise automatically with monthly spend
- Fine-tuning with one-click deployment of the tuned model
- Reserved and elastic GPU capacity for steady production workloads
- Private and bring-your-own-cloud deployment for enterprises
SiliconFlow is one of the simplest ways to call Chinese open-weight models from existing code: one OpenAI-compatible endpoint, more than 170 models, per-token prices, a handful of free small models and $1 of starter credit on the international site. Rate limits climb with monthly spend, and reserved GPUs, fine-tuning and private deployment cover heavier use.
Its June 2026 Hong Kong prospectus shows fast growth, heavy losses and a funding headline larger than the filing records, so weigh vendor stability and get uptime terms in writing. ASEAN teams can contract with the Singapore entity behind siliconflow.com.
What we found
SiliconFlow (硅基流动) is a Beijing company that sells API access to other developers' AI models rather than training its own. Yuan Jinhui, a former Microsoft Research Asia researcher who created the OneFlow deep-learning framework, set it up in August 2023, and its model platform went live in May 2024 under the name SiliconCloud. It now runs two storefronts: siliconflow.cn for China, billed in yuan, and siliconflow.com, operated by SiliconFlow Labs Pte. Ltd. in Singapore under Singapore's data-protection law and billed in US dollars.
Both sites put DeepSeek, Qwen, GLM, Kimi, Tencent Hunyuan and other models behind one OpenAI-compatible API, alongside image, video, speech and embedding models. The company's own figure is more than 170 models; earlier versions of this page said 200-plus, which nothing on its sites supports. Prices are per million tokens, for example $0.15 in and $0.60 out for DeepSeek-V4.1-Flash on the international site, and the Chinese site charges some models less overnight. A few small models are free, including Qwen2.5-7B-Instruct, Hunyuan-MT-7B and the BAAI embedding models. International accounts get $1 of credit, and rate limits rise with monthly spend.
The money story changed in mid-2026. On 30 June SiliconFlow filed to list on the Hong Kong Stock Exchange under Chapter 18C, the route for specialist-technology companies not yet earning enough to qualify as commercial. Its prospectus, as reported by The Paper, shows 2025 revenue of ¥55.3 million against a loss of ¥345 million, a negative gross margin driven partly by free-token vouchers, and a ¥7.74 billion valuation after a June B+ round. The company had announced a B round of more than ¥2 billion two weeks earlier; Sina Finance pointed out that the filing records ¥1.26 billion across B and B+. A further B+ close and a C round followed in September.
For developers who want Chinese open models without running GPUs, it is cheap and quick to wire in. Teams that need contractual uptime should get the reserved-instance terms in writing.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including SiliconFlow 硅基流动's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with SiliconFlow 硅基流动 unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to SiliconFlow 硅基流动 directly →
Spotted something out of date? Suggest an update →
Alternatives to SiliconFlow 硅基流动
More in LLMs & Chat
