If you already work in Cursor, try OpenAny through a compatible custom-model connection before changing your editor or deploying a gateway. You can compare model routes and fund model and external API usage through the same OpenAny balance.
This is not a promise that every model or Cursor mode accepts an arbitrary endpoint. The useful first outcome is a verified connection for one chat workflow, with a clear rollback if your Cursor version rejects the route.
Connection values
| Setting | Value or decision |
|---|---|
| Provider interface | OpenAI-compatible custom-key connection |
| Base URL | https://api.openany.si/v1 |
| API key | Your authorized OpenAny key |
| Model | Exact catalog ID supported by the selected Cursor workflow |
| Initial example to inspect | qwen3.7-max, not a certification of Cursor compatibility |
The reviewed OpenAny Qwen template uses /v1/chat/completions. If Cursor's selected mode needs Responses instead, that template does not establish support. Resolve the protocol mismatch before paying for a test.
Set up one model first
- Open Cursor Settings > Models and record the existing provider configuration without recording secret values.
- Find the OpenAI-compatible key setting. In versions exposing Override OpenAI Base URL, enter the base above and use your OpenAny key.
- Add or enable the exact model ID if that version supports custom models. Avoid Auto mode for the first test: you need to know which model is selected.
- Save the configuration and check the requested destination and model. Do not assume a successful provider-key verification has tested the custom model; the verification request may use a different model.
- With an approved budget, send a minimal chat prompt. Only then test streaming or the specific agent/tool mode you plan to use.
The override can affect more than one model rather than being a per-model setting. Cursor's support forum documented this behavior in 2025; check your installed version before mixing native OpenAI and gateway models. Do not leave unrelated models enabled and assume they still use their previous destination.
Make the first test useful
Use a disposable text file, not a private production repository. Ask for a two-sentence explanation, then a small read-only context task. Verify the response, selected route and OpenAny usage entry. Do not interpret a model's own claim about its identity as routing proof.
| Failure | Action |
|---|---|
| Key accepted but custom model rejected | Check model allowlisting and which model key verification used |
| Chat works but another mode fails | Check that mode's protocol and feature requirements |
| Native models unexpectedly fail | Restore the global Base URL override |
| Repeated retry or tool loop | Stop the session and inspect usage before another attempt |
The test is complete only when the intended workflow works and its charge is visible. No paid result or latency benchmark is supplied here.
Know which bill changes
Gateway-routed inference uses OpenAny's applicable model rate. This does not cancel a Cursor subscription or move Tab completion to OpenAny. Cursor documents separate plan behavior for BYOK, including possible Teams/Enterprise token-rate charges. Compare the complete workflow bill, not just one API rate.
For tool access without switching Cursor's main model, use the OpenAny agent setup. A skill can add data and model calls as tools, but it is not a Base URL override and should not be presented as one.
Use with your Agent
The copied task prepares a version-aware setup; it does not execute a paid request.
Make this example yours
Check whether this Cursor version can use qwen3.7-max through OpenAny, then prepare a reversible setup.
Task preview
Read https://openany.si/AGENT.md and follow its setup and credential-safety instructions. Task: Prepare a Cursor connection to OpenAny for qwen3.7-max. Check the installed Cursor version with user permission, its current custom-key and Base URL settings, official documentation, and the selected OpenAny route's public request template. Distinguish Chat Completions from Responses and do not promise all Cursor modes support arbitrary models. Reuse only explicitly authorized credential configuration; never request a key in chat or scan files for secrets. Explain global override effects and Cursor plan charges. Preserve prior settings and return a step-by-step proposed change, rollback, and a budget-gated test plan. Do not edit configuration without approval, make paid requests, top up, or claim that OpenAny changes Tab completion. Guide: https://openany.si/guides/cursor-openany-base-url/ Ask for a budget and approval before billable requests. Never request credentials in chat. Do not substitute providers or execute other actions without approval. Preserve sources and timestamps; report missing data and actual results honestly.
Browse OpenAny models, create an account, or compare Qwen overseas pricing before choosing the route.
Sources
Frequently asked questions
What Base URL should I use?
For a verified OpenAI-compatible route, use https://api.openany.si/v1 as the base, not the full chat/completions URL.
Does this replace Cursor Tab completion?
No. Cursor documents custom keys for chat models; Tab completion uses Cursor's built-in models.
Does a custom key remove every Cursor charge?
No. Cursor plan and feature charges remain separate; Teams and Enterprise may apply additional token-rate charges.