We may earn a commission if you buy through links on this page. It never changes a score or a ranking. How we make money
Pricing is quote-based.
SiliconFlow's listed pricing has no free plan or stated free-trial allowance, though the homepage invites users to get started for free without specifying included credit. Serverless inference is postpaid and pay-as-you-go, with rates set by model and billing unit. For example, the listed Qwen3.5-9B rate is $0.10 per 1M input tokens and $0.15 per 1M output tokens. Image and video models use per-item rates. GPU and fine-tuning charges need separate confirmation.
01Every plan, line by line
| Plan | Billed annually | Billed monthly | Built for |
|---|---|---|---|
| Serverless usage-based | — | Quote | Teams paying for model inference |
02What the plan grid doesn't tell you
Model mix
Token rates vary by model and split input from output. Estimate spend using your expected model mix and billable token volumes.
Image and video usage
Listed image and video models use per-image or per-video rates rather than the token pricing used for language models.
GPU capacity
Reserved and elastic GPUs are listed, but their prices are not included with the serverless model rates. Request a quote for your deployment.
Fine-tuning charges
Fine-tuning and deployment are listed as products, but their rates are not stated alongside the serverless examples. Confirm charges before training.
03Which plan should you buy?
| If you are… | Get | Why |
|---|---|---|
| Prototype using hosted inference | Serverless usage-based | Pay per model usage without a fixed monthly subscription price; no free-plan allowance is listed. |
| Production app with variable traffic | Serverless usage-based | The postpaid model scales with usage. Track token and media consumption against each model's rate. |
| Team needing reserved capacity | Serverless usage-based | Reserved GPUs are listed separately; ask SiliconFlow for pricing and commitment terms because serverless rates do not cover them. |
04Is SiliconFlow worth the money?
SiliconFlow's pay-as-you-go model is useful when you want to select models without committing to a fixed monthly subscription. The listed rates are concrete for several token and media models, but total spend depends on model choice and actual usage. There is no free plan or stated trial allowance to use as a dependable production budget. Ask for separate prices for reserved or elastic GPUs and fine-tuning, then set internal usage alerts and confirm how account rate limits affect your workload.
SiliconFlow pricing FAQ
What is the cheapest listed SiliconFlow model rate?
Among the published examples, Qwen3.5-9B is listed at $0.10 per 1M input tokens and $0.15 per 1M output tokens. That is not a universal starting price: rates depend on the selected model and billing unit.
Does SiliconFlow charge a monthly subscription?
The pricing details describe postpaid, pay-as-you-go billing and show no fixed monthly subscription price. Reserved GPU billing may have separate terms, so ask SiliconFlow for a quote if you need dedicated capacity.
How much does SiliconFlow charge for image and video generation?
The listed examples are FLUX.1-dev at $0.014 per image and Wan2.2-T2V-A14B at $0.29 per video. Check the current model pricing page for the exact model you plan to call, since rates vary.
Are fine-tuning and GPU costs included in serverless pricing?
The serverless examples provide model inference rates, not a price for fine-tuning or reserved and elastic GPUs. SiliconFlow lists those products separately. Request their applicable rates and billing terms before estimating a combined deployment budget.