Start
Pricing and billing
You pay per token, from credits you buy. $5 is about 2,700+ replies on DeepSeek or 6,500+ replies on GLM, at 3,000 tokens in and 250 out.
Checked 26 October 2026
Prices
| Model | Per 1M tokens in | Per 1M tokens out | One reply | $5 buys | $25 buys |
|---|---|---|---|---|---|
| DeepSeek V4.1 Flash Uncensored | $0.45 | $1.75 | $0.0018 | 2,700+ replies | 13,000+ replies |
| GLM 5.3 Flash Uncensored | $0.20 | $0.65 | $0.0008 | 6,500+ replies | 32,000+ replies |
A reply here is 3,000 tokens in and 250 out, thinking off. 1,000 credits are $1.
Work out your own
GLM always thinks briefly, which adds output tokens this leaves out.
Credits
- 1,000 credits are $1. Buy them in the console, from $5, in your own currency where the checkout offers it. A first top-up can be up to 100,000 credits.
- Credits you buy never expire, and the app spends them too. Credits are non-refundable.
- The API spends bought credits only. A plan’s monthly credits stay in the app.
How a call is billed
- Before the model is asked, the call holds the most it could cost: its prompt and its
max_tokensat the model’s price. - When the reply ends, the hold settles to the tokens the model really used, from
usage. - The bill is worked out to a millionth of a credit and charged in whole credits. The fraction carries to your next call, so a short call isn’t rounded up.
| The call | Costs |
|---|---|
| Answered | Its tokens in and out, thinking included |
Refused by us (400, 402, 422, 429) | Nothing |
Failed before a word (502, 503) | Nothing |
| Cut off mid-stream by the model | What it wrote, nothing more |
| Stopped by you mid-stream | What was sent before you stopped |
| Made with the test key | Nothing |
See what you spent
Every reply says what it cost: usage.nokeep_credits in the body, and x-nokeep-credits and x-nokeep-balance headers on a whole reply. The console shows spend by day and every call, and exports the last 30 days as a CSV.
Questions
Is there a free tier?
There is a free test key that checks your setup without reaching a model. Real replies start at $5 of credits.
Why does thinking cost more?
Thinking is written as output tokens, at the output price. DeepSeek thinks only when you ask; GLM always thinks briefly.
Do pictures cost extra?
A picture counts as input tokens, as the model reads it. There is no separate charge.