AI credit rates
What one AI request actually costs.
Lovable does not publish a fixed rate per request. So the atlas sent real requests and read the cost of each one from the AI Gateway's own log. Measured 2026-10-17.
The test
Summarize a 220-token note in three sentences, three times per model, streamed, with low reasoning effort where the model supports it. Each figure below is a logged cost, not an estimate.
| Model | Credits per request | Per 1,000 requests | Logged runs |
|---|---|---|---|
| Gemini 3.1 Flash Litegoogle/gemini-3.1-flash-lite | 0.0003185 | 0.319 | 0.0003285, 0.0003045, 0.0003225 |
| Gemini 3.8 Flashgoogle/gemini-3.8-flash | 0.0037335 | 3.734 | 0.003936, 0.004041, 0.0032235 |
| GPT-6 Astra (Lovable's default)openai/gpt-6-astra | 0.0111067 | 11.107 | 0.01144, 0.01094, 0.01094 |
How to read this
The cheapest model here cost about 35 times less than Lovable's default for the same short summary. Gemini 3.8 Flash wrote much longer answers, which is most of its cost.
Longer notes, longer answers, images or more reasoning all cost more. For your own app, open Cloud, then AI, and read a real request. Then put it in the AI Gateway calculator.
One task on one day. Lovable can change rates at any time. What a credit costs in dollars depends on your plan. The atlas is independent and not affiliated with Lovable AB.