Cursor charges a markup of i think 20%, so the cheapest option would be to use the api key or use claude code inside cursor ( i think you have a anthropic subscription, so there you have more free credits to spend )
You can keep using premium models after you run out of credits in two ways:
Use free models that sometimes appear under bonus usage (for example, Grok Code). Otherwise, enable On-Demand usage.
For early renewal or restarting the Ultra plan before your billing cycle, please contact [email protected]. Billing changes are handled there.
On-Demand pricing is the same as usage-based billing shown in the billing page.
Using your own Claude Opus 4.5 API key is usually not cheaper. With Ultra, the $200 plan gives about $400 worth of API-token value plus bonus usage. With On-Demand or your own API key, you pay as you go (so $200 spend equals $200 value).
Takeaway to avoid burning tokens too fast:
Use Opus for planning, better yet. Try other models such as GPT 5.2 High / GPT5.2 xHigh
Use Composer-1, Sonnet 4 or Cheapest for execution
New Task, or different task, make new chat to avoid Context (more context, more token)
Check MCP, make sure only use MCP that u need, since more MCP, more tools and more token
Hi all,
Thank you so much for quick response. I contacted the Cursor support team and I only have two options: 1) using on-demand and 2) using the auto mode.
Have anyone try to use 2 accounts sequence, like A ā B ā A to reset account renew earlier?
Best
The community answers above cover your questions well! To summarize:
Continue using premium models: Enable On-Demand usage, or use your own API key (BYOK)
On-Demand pricing: Same API rates as your plan usage (model pricing)
The token-saving tips from @Naufaldi_Rafif are solid - using lighter models for execution and starting new chats for new tasks can help credits last longer.
One thing Iād add is using subagents to explore codebase, etc.