The buying decision
Judge the provider on accepted repository edits and total repair turns. A low displayed token cost is not enough if Aider cannot reliably apply the result.
What makes an Aider session expensive
Repository context, requested output, model choice and repeated edits all affect usage. Start with the files needed for one change and a concrete acceptance command. Include failed patches and repair attempts in your comparison. Keep the same repository revision and requested outcome when comparing two providers; otherwise you are measuring different work.
Connect an OpenAI-compatible endpoint
Aider's documented configuration uses OPENAI_API_BASE and OPENAI_API_KEY, followed by --model openai/<model-name>. For Standard Compute the base is https://api.stdcmpt.com/v1 and the model argument is openai/standardcompute. Set the key in your local environment. Keep existing Aider configuration available so that this trial can be reversed.
Check model metadata and edit behavior
A custom or routed model may not have a matching entry in the client's built-in metadata. Check context limits, edit format and token accounting before trusting a long session. A client estimate using generic or missing pricing is not your provider invoice. Confirm an actual file change, inspect the diff and run the same test suite before judging the model connection usable.
When a subscription makes sense
A fixed monthly API bill can fit regular work when the included allowance covers the tasks you actually complete. Direct metered use may cost less in quiet months and makes a specific model easier to compare. Standard Compute offers a fixed price and finite compute allowance; it does not guarantee an unlimited number of Aider tasks. Keep the subscription fee separate from per-task allowance consumption in your calculation.
Run your own comparison
Use the same task, starting revision and acceptance check. Record every retry and failed attempt so the result reflects the work you actually paid for.