Our take
SiliconFlow is a China-headquartered host for downloadable models with 33 tracked offers and measured throughput data. It is one of the cheaper options for several models and offers a zero-retention setting, though key compliance disclosures remain unverified.
Use this for budget-conscious inference on Chinese and open-weight models when a zero-retention option is required. Pick it when throughput needs are modest and price matters more than peak speed. Skip it if EU data residency or SOC 2 attestation are hard requirements, or if China-headquartered hosting raises data-sovereignty concerns for your organisation.
- Among the cheapest tracked options for several models — GPT-OSS Safeguard 20B, Qwen3-5-9B and Qwen3 Coder 30B A3B are all priced at the low end of their ranges.
- Explicit zero-retention option available, and it does not train on customer prompts.
- Throughput data is available for every tracked offer, from 6 TPS to 32 TPS across the sample set.
- Compliance and geographic disclosures are thin: SOC 2 and EU endpoint are both unverified in our data.
- Headquartered in China, which may raise data-sovereignty concerns for some buyers.
- Throughput on larger models lags smaller ones: a 5.3× gap between the slowest and fastest measured offers, and a 1.5× gap even within the same model family.
Privacy & data handling
- ✓
- Yes
- ✗
- No
- No answer on record
The colour says whether the answer favours you, not whether it is a yes — not training on your prompts earns a green cross, no zero-retention option earns a red one.
Prompt retention: none.