How routing works
How evolved.to reaches OpenAI and Anthropic, routes simple prompts to lighter models and scales effort with your task.
What evolved.to is
evolved.to is an independent API gateway. It is not ChatGPT or Claude, and it is not an official product of OpenAI or Anthropic. It gives you one key for the Claude and ChatGPT model families and handles the connection to the providers for you. Our prices are possible because of how that connection is engineered, which this page explains.
The proxy layer
Every request goes through our proxy infrastructure and on to OpenAI's and Anthropic's servers, and the response comes back in the same protocol you called. Requests from all customers share that infrastructure, which lets us use provider capacity far more efficiently than a single account could, and pass the difference on in our credit packs.
Smart routing
Not every prompt needs a frontier model. When a request is short and simple, such as a rename, a quick question or a one-line fix, the router can send it to a lighter model from the same family: Claude requests stay on Claude models and ChatGPT requests stay on GPT models, so the style and format of answers stay consistent. Anything substantial, such as multi-file changes, long context or real reasoning, runs on the model you named.
| Example prompt | You asked for | Served by |
|---|---|---|
| Rename getUser to fetchUser | claude-opus-5 | claude-haiku-4.5 |
| What does this regex match? | gpt-astra-6 | gpt-5-mini |
| Add tests for the orders API | claude-opus-5 | claude-opus-5 |
| Refactor the auth module across 12 files | gpt-astra-6 | gpt-astra-6 |
Adaptive effort
Models are not run at 100% of their capacity on every call. Reasoning depth, maximum output length and the amount of context sent scale with the size of the task, and reach the model's maximum on large, complex projects. Simple work stays fast and cheap without holding back on hard problems.
Small edit or quick question
Light
Short reasoning, compact answers, only the context the prompt needs.
Feature work
Medium
More reasoning and room for longer, multi-step answers.
Multi-file changes
High
Deep reasoning across the files and history involved.
Large projects
Maximum
The model's full depth, output length and context.
Billing and transparency
- Each request is billed at the rate of the model that actually served it.
- The
modelfield in every response names the model that served the request. - Rates for every model are public on the Models page, and the full terms are in our Terms of Service.