Overview
SiliconFlow is a domestic AI inference platform known as 'the lowest price on the internet', offering API services for mainstream models such as DeepSeek, Qwen, and Llama. With domestic computing power and an optimized inference engine, it significantly reduces call costs while maintaining performance, making it a popular choice for domestic developers using open-source models.
The platform continues to expand its model coverage. As of 2026-09-14, it added the Tencent Hunyuan Hy4 preview (770B total parameters, 49B activated, 1M context, Apache 2.0), priced at $0.834 input and $2.501 output per million tokens, and now covers APIs for the DeepSeek, Qwen, and Hunyuan families among others. High-speed versions have also been rolling out, including Kimi K2.7 Code (2026-06-18), DeepSeek-V4-Flash-0731 (2026-08-26), and GLM-5.3 (2026-09-04).
Key Features
- Lowest Price API on the Internet: API prices for mainstream models are lower than official and other platforms.
- Multi-Model Aggregation: One-stop access to mainstream models such as DeepSeek/Qwen/Llama/Mistral, with newly added high-speed versions like GLM-5.2, Kimi K2.7, and DeepSeek-V4-Pro Flash. The platform now covers APIs for the DeepSeek, Qwen, and Hunyuan families among others, and has added the Tencent Hunyuan Hy4 preview (770B total parameters, 49B activated, 1M context, Apache 2.0) as of 2026-09-14.
- Domestic Computing Power: Deployed in domestic data centers, offering low latency and compliance without worries.
- OpenAI Format Compatibility: API interface is compatible with OpenAI SDK, with zero migration cost.
- Free Credits: New users receive free Token credits upon registration.
- High-Speed Inference: Language model speed increased by more than 10 times, image generation in 1 second, and voice generation in 100 milliseconds. High-speed model versions continue to expand, including Kimi K2.7 Code (2026-06-18), DeepSeek-V4-Flash-0731 (2026-08-26), and GLM-5.3 (2026-09-04).
- High Stability: Provides comprehensive monitoring and fault tolerance mechanisms, with professional technical support to ensure high availability.
- High Security: Supports BYOC deployment with isolated computing, networking, and storage, meeting industry standards and compliance requirements.
- Reserved Instances: A one-stop solution for enterprise core inference scenarios, providing exclusive computing power, precision assurance, and cost optimization.
- Private Deployment: Offers enterprise-level private deployment solutions, covering model performance optimization, deployment, and operation and maintenance in one go.
Use Cases
- Domestic developers calling open-source large models
- Low-cost inference backend for AI applications
- Model effect comparison and selection
- Domestic deployment of enterprise-level AI applications
- AI projects for individual developers
- Enterprise core inference scenarios (reserved instances)
- Private deployment to meet diverse scenario needs
Pros
- Lowest price on the internet, developer's first choice
- Domestic computing power, low latency + compliance
- OpenAI format compatibility, zero migration cost
- Multi-model aggregation, all in one platform
- High-speed inference, language model speed increased by more than 10 times
- High stability, enterprise-level SLA guarantee
- High security, supports BYOC deployment
Pricing
Billed by Token, prices are far lower than the official APIs of each model. New users receive free credits upon registration. Reserved instances offer a better cost structure. Specific prices vary by model. For reference, the Tencent Hunyuan Hy4 preview added on 2026-09-14 is priced at $0.834 input and $2.501 output per million tokens.
Summary
SiliconFlow is the best value choice for domestic developers using open-source models, suitable for AI projects that are cost-sensitive and require domestic compliant deployment. Its continuously expanding model lineup, now spanning DeepSeek, Qwen, Hunyuan, and other families with high-speed variants, keeps it a strong fit for cost-effective inference on domestic compute.
Version History
- The official website shows that SiliconFlow has launched the high-speed version of Kimi K2.7 Code (2026-06-18) and the high-speed version of GLM-5.3 (2026-09. (2026-09-22): The official website shows that SiliconFlow has launched the high-speed version of Kimi K2.7 Code (2026-06-18), the high-speed version of GLM-5.3 (2026-09-04), the high-speed version of DeepSeek-V4-Flash-0731 (2026-08-26), and other models, as well as multiple corporate events, but the review did not mention these new developments.
- Adds Tencent Hunyuan Hy4 preview to the platform (2026-09-14): SiliconFlow added Tencent Hunyuan Hy4 preview (770B total parameters, 49B activated, 1M context, Apache 2.0), priced at $0.834 input and $2.501 output per million tokens. The platform positions itself as cost-effective inference on domestic compute, currently covering APIs for the DeepSeek, Qwen and Hunyuan families among others.