GPT, Claude and Gemini API pricing
Prices per 1M tokens in USD, current as of ноябрь 2026. One balance covers every model: you pay only for the tokens you use.
OpenAI GPT 8 models
| Model | Input, $/1M | Output, $/1M |
|---|---|---|
gpt-6-astra |
$4.00 | $20.00 |
gpt-5.6-sol |
$1.60 | $8.00 |
gpt-5.6-terra |
$0.80 | $4.80 |
gpt-6-sol |
$0.80 | $4.00 |
gpt-6.1-sol |
$0.80 | $4.00 |
gpt-6.1-terra |
$0.40 | $2.00 |
gpt-5.6-luna |
$0.08 | $0.48 |
gpt-6-luna |
$0.04 | $0.20 |
Google Gemini 3 models
| Model | Input, $/1M | Output, $/1M |
|---|---|---|
gemini-3.5-flash |
$0.60 | $3.60 |
gemini-3.8-flash |
$0.33 | $1.65 |
gemini-3-flash |
$0.20 | $1.20 |
Other models 4 models
| Model | Input, $/1M | Output, $/1M |
|---|---|---|
glm-5.3 |
$0.929 | $1.46 |
deepseek-v4-pro |
$0.38 | $0.76 |
deepseek-v4-flash |
$0.09 | $0.18 |
glm-5.3-flash |
$0.099 | $0.166 |
How billing works
- You top up one balance and use any model from the list: there is no subscription and no per-model commitment.
- Each request is charged by the tokens it actually used: input and output tokens, plus reasoning and cache tokens where the model reports them.
- The price list above is the one the API applies, and the same model ids are used in your requests.
Connect in one line
The API works with OpenAI, Anthropic and Gemini clients. Point your tool at the base URL and use your key.
curl https://llm-router.org/v1/chat/completions \
-H "Authorization: Bearer llmr_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "gpt-6-astra", "messages": [{"role": "user", "content": "Hello"}]}'Replace llmr_YOUR_KEY with the key from your dashboard. See the docs for Claude Code, Codex and other tools. Read the docs →
Questions about pricing
What are the prices per 1M tokens?
The table lists the input and output price per 1M tokens for every available model, in US dollars.
Is there a subscription or a minimum commitment?
No. You top up the balance and it is charged as you use it, across all models.
Which API formats are supported?
OpenAI-compatible chat completions and responses, the Anthropic Messages API and Gemini-compatible requests, so the usual SDKs and tools work after changing the base URL.
Do you store my prompts?
We log metadata only: model, tokens and latency. Prompt content is never stored and never trained on.