Pay-as-you-goPay only for what you use
$0No charge for failed tasks
NoneHidden fees
14 daysLogin session validity
| Model | Type | Input / 1M | Output / 1M | Cache R/W | You save | Image 1K | Image 2K | Image 4K |
|---|
| claude-sonnet-4-6 |
Chat |
$1.5
List $3.0 |
$7.5
List $15.0 |
$0.15
wr $1.875 |
↓50% |
— |
— |
— |
| gemini-2.5-flash |
Chat |
$0.3
List $0.3 |
$1.2
List $2.5 |
$0.03
wr $0.375 |
↓52% |
— |
— |
— |
| gemini-3-flash-preview |
Chat |
$0.3
List $0.5 |
$1.2
List $3.0 |
$0.03
wr $0.375 |
↓60% |
— |
— |
— |
| gemini-3.1-pro |
Chat |
$0.5
List $2.0 |
$3.0
List $12.0 |
$0.05
wr $0.625 |
↓75% |
— |
— |
— |
| gemini-3.5-flash |
Chat |
$0.45
List $1.5 |
$2.7
List $9.0 |
$0.045
wr $0.562 |
↓70% |
— |
— |
— |
| gemini-3.6-flash |
Chat |
$0.6 |
$3.6 |
$0.06
wr $0.75 |
— |
— |
— |
— |
| gemini-3.6-flash-high |
Chat |
$0.6 |
$3.6 |
$0.06
wr $0.75 |
— |
— |
— |
— |
| gemini-3.6-flash-low |
Chat |
$0.6 |
$3.6 |
$0.06
wr $0.75 |
— |
— |
— |
— |
| gemini-3.6-flash-tiered |
Chat |
$0.6 |
$3.6 |
$0.06
wr $0.75 |
— |
— |
— |
— |
| gpt-5.5 |
Chat |
$0.7
List $5.0 |
$4.2
List $30.0 |
$0.07
wr $0.875 |
↓86% |
— |
— |
— |
| gpt-5.6-luna |
Chat |
$0.8
List $2.5 |
$4.8
List $15.0 |
$0.08
wr $1.0 |
↓68% |
— |
— |
— |
| gpt-5.6-sol |
Chat |
$2.0
List $8.0 |
$12.0
List $48.0 |
$0.2
wr $2.5 |
↓75% |
— |
— |
— |
| gpt-5.6-terra |
Chat |
$1.2
List $5.0 |
$7.0
List $30.0 |
$0.12
wr $1.5 |
↓77% |
— |
— |
— |
| gpt-image-2 |
Image |
— |
— |
— |
— |
$0.015 |
$0.025 |
$0.04 |
| nano-banana-2 |
Image |
— |
— |
— |
— |
$0.025 |
$0.0375 |
$0.0625 |
| nano-banana-pro |
Image |
— |
— |
— |
— |
$0.04 |
$0.06 |
$0.1 |
| veo-3.1 |
Chat |
— |
— |
— |
— |
— |
— |
— |
Prices are in USD. Token charges are based on the usage reported by the upstream provider; images are billed by the number generated and their resolution.
Caching: cached input (cache read) is billed at 1/10 (0.1×) of the input rate, cache writes at 1.25×. A hit needs a stable prefix long enough (Claude Opus ≥ 4096 tokens, Sonnet ≥ 2048) repeated within a short window — e.g. Claude Code with a large system prompt over a codebase hits every turn, cutting input cost to 1/10. Short Playground chats are too small to cache.
