Pricing
Pay as you go, from your balance
Top up your balance, then pay only for what you use, per token for text — no subscription, no hidden fees.
Starter
$2.78
$2.78 balance
To get started
- Access to all active models
- Web chat + API
- No subscription
Pro
Popular$8.33
$8.83 balance
+6% bonus — regular use
- Everything in Starter
- +6% balance bonus
- Great for Claude Code
Business
$27.78
$31.94 balance
+15% bonus — high volume
- Everything in Pro
- +15% balance bonus
- For teams / production
Or top up a custom amount from a low minimum. Usage is billed per token at each model's rate.
Top up any amount from a low minimum. Balance bonuses: +6% on larger top-ups, +15% on the highest tier.
Savings calculator
Estimate your monthly cost from our live published rates. No subscription — you pay only for the tokens you actually use.
A request ≈ one message in + one reply out.
Estimated / month
$1.06
Per request: ≈ $0.0004
- Pay with QRIS — no credit card needed
- Billed in Rupiah, no FX markup
- No subscription, no lock-in
Estimates from live rates. Actual cost follows the real token count of each request — never more.
Worked examples
Costs are computed straight from each model's rate. Here are estimates for a few scenarios.
| Model | Tokens (in / out) | Estimated cost |
|---|---|---|
Claude Opus 4.8 Anthropic · One short message | 1,000 / 1,000 | ≈ $0.012 |
Estimates are rounded. Your actual cost follows the real token count of each request.
Rates by model
Per-token rates for each text model — input and output billed separately.
Text
| Model | Rate |
|---|---|
GPT-5 MiniVision OpenAI | Input: $0.131 / 1M tokens Output: $0.261 / 1M tokens |
Claude Opus 4.8Vision Anthropic | Input: $1.95 / 1M tokens Output: $9.75 / 1M tokens |
How billing works
Cost = token count × the model's rate. Input tokens come from the messages you send, output tokens from the length of the reply. Charged against your balance.
Cache tokens (Claude)
For Claude models, cache tokens are billed separately: cache reads are cheaper and cache writes are slightly more than the standard input rate. Other models don't break out cache tokens.