·Partnership Announcement·中文版本 →

TheRouter × SiliconFlow: China-Region AI Models at 40–80% Lower Cost

We're partnering with SiliconFlow to bring DeepSeek R1, DeepSeek V3, and Qwen3 series models to TheRouter via Aliyun k3s — giving developers China-region inference at a fraction of the cost of US-based providers.


TheRouter announces a partnership with SiliconFlow to provide China-region AI inference. SiliconFlow routes DeepSeek R1, DeepSeek V3.2, and Qwen3 models through Aliyun k3s. Cost savings compared to Bedrock: DeepSeek R1 at $0.55/$2.19 per MTok versus $3.50/$17.50, DeepSeek V3 at $0.14/$0.28 versus $0.90/$2.70. No code changes required — swap the model name and TheRouter handles routing automatically with Bedrock as primary, SiliconFlow as fallback.

Why SiliconFlow?

SiliconFlow operates a high-throughput inference platform on Alibaba Cloud, purpose-built for China-region developers and teams that need low-latency access to open-weight models. Their Pro-tier API supports up to 5,000 RPM per model — fast enough for production workloads.

For teams already using TheRouter with DeepSeek or Qwen models via AWS Bedrock, SiliconFlow provides an automatic fallback path. When Bedrock is unavailable or slow, traffic transparently shifts to SiliconFlow without any code changes.

Cost Comparison

SiliconFlow pricing for the same models available on Bedrock:

ModelSiliconFlow InputSiliconFlow OutputSavings vs Bedrock
deepseek/deepseek-r1$0.55/MTok$2.19/MTok~84% cheaper
deepseek/deepseek-v3.2$0.14/MTok$0.28/MTok~85% cheaper
qwen/qwen3-235b$0.14/MTok$0.55/MTok~70% cheaper
qwen/qwen3-32b$0.03/MTok$0.07/MTok~75% cheaper

Savings calculated against AWS Bedrock on-demand pricing as of March 2026.

How It Works

SiliconFlow runs as a priority-1 provider in TheRouter's routing engine — Bedrock stays as the primary (priority 0) for US-based teams, while SiliconFlow activates as the fallback or when a China-optimized routing preference is set.

No changes to your existing code. Just use the same standard model name:

# Works exactly the same — TheRouter picks the best provider
curl https://api.therouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $THE_ROUTER_API_KEY" \
  -d '{
    "model": "deepseek/deepseek-r1",
    "messages": [{"role": "user", "content": "Explain quantum entanglement"}]
  }'

TheRouter normalizes the response — including preserving DeepSeek R1's reasoning_content field and normalizing finish reasons — so the response looks the same regardless of which provider actually served the request.

Available Models

The following models are now available through TheRouter via SiliconFlow:

View the full model catalog →

Getting Started

If you're already using TheRouter, SiliconFlow is automatically available as a fallback for covered models. No changes needed.

New to TheRouter? Sign up and start routing DeepSeek and Qwen models in minutes:


Questions? Reach out on GitHub or check the integration docs.