Hard facts side by side: free tiers, prices, limits, SDKs and privacy. Every value links to the official page it came from and is re-checked every week.
| Fact |  Replicate |  fal |
|---|---|---|
| Free tier | Yes Selected Try for Free models run a limited number of times without a card. Source, Sep 29, 2026 | Not listed |
| Starting price | $0.000025 per second CPU (Small) hardware, $0.09/hr; many models are priced per output instead. Source, Sep 29, 2026 | $0.02 per megapixel Qwen image model; GPU compute from $1.10/hr. Source, Sep 29, 2026 |
| Rate limit | 600 predictions per minute Accounts with credit but no payment method: 6 requests per minute. |
| 2 concurrent requests New accounts; scales up to 40 with credit purchases. Source, Sep 29, 2026 |
| Size limit | 100 MB per input file Local file uploads via client; data URIs recommended only under 1 MB. Source, Sep 29, 2026 | Not listed |
|---|
| Storage | Files kept 1 hour API prediction inputs, outputs, files and logs are removed after an hour by default. Source, Sep 29, 2026 | Request data kept 30 days JSON inputs/outputs stored 30 days by default; CDN media expiration is configurable. Source, Sep 29, 2026 |
|---|
| Official SDKs | PythonJavaScript / TypeScriptGoSwiftSource, Sep 29, 2026 | PythonJavaScript / TypeScript |
|---|
| Open source | Partly Cog, the model packaging tool, is open source; the platform is not. Source, Sep 29, 2026 | Not listed |
|---|
Wirestack