Free AI API models
Eligible accounts get 3,000,000 free tokens per month, shared across eligible :free models.
3,000,000 tokens per account per month
Input and output tokens use the same pool across every eligible free model and API key on your account. This is not a separate allowance for each model. It resets at the start of each calendar month, 00:00 UTC. Unused monthly tokens do not roll over.
Sign up and verify your email before using the allowance. No phone verification is required. Requests remain subject to model availability, context limits and your API key's rate and usage limits. The free program currently allows at most 16,384 output tokens per request; a lower key or model limit still applies.
Published free models
| Model | API model ID | Context | Published capabilities |
|---|---|---|---|
| DeepSeek V4.1 Flash 1017 Free | deepseek-v4.1-flash-1017:free | 1,000,000 | text, vision, cache, tools, stream |
| GPT 6 Luna Free | gpt-6-luna:free | 272,000 | text, vision, cache, tools, stream |
| Qwen3.8 Flash Free | qwen3.8-flash:free | 1,000,000 | text, vision, cache, tools, stream |
When the remaining free tokens cannot cover a request
Suppose you have 1,000 free tokens available and a new request needs a 1,500-token reservation. The free pool cannot cover the request, even though it still has tokens. If both fallback settings are on and the wallet can cover the required reservation, the request can use the paid route's rates. With either setting off or insufficient wallet credit, the request stops. The shared pool resets at 00:00 UTC on the first day of the month.
A token allowance counts usage against one pool of eligible aliases. Wallet credit pays the published dollar rate. Fallback changes the funding source for the request. It does not pick a different model for you. For a published paid counterpart, see the GPT 6 Luna rates.
| Account and key settings | Wallet balance | Result |
|---|---|---|
| Fallback enabled on both | Sufficient for the required reservation | The request can use the corresponding paid model's published wallet rate. |
| Fallback disabled on either | Any amount | The free request stops. Enable fallback deliberately or select a paid model. |
| Fallback enabled on both | Empty or insufficient | The request stops. Add funds before using wallet billing. |
A paid model ID uses its own billing rules. A :free suffix is an alias with allowance rules, not unlimited inference or an independently priced model. In-flight requests reserve allowance, so available tokens can be lower than tokens not yet finally used. A request whose required reservation exceeds the remaining pool can require fallback even before the counter reaches zero.
Make your first request
Create an account, verify your email, create an API key and choose an exact eligible ID above. Use the Chat Completions quickstart. Review the account fallback preference and the setting on each key before enabling wallet charges.
For a paid workload, use the token cost calculator. Compare DeepSeek and Qwen rates or read how input and output affect your bill. Baca dalam Bahasa Indonesia.