Cline is using too many tokens / API costs too high
Cline reads whole files into context and loops over tool results, so per-token bills of $60–200+/month are normal for daily use. The tactical fixes are context discipline and cheaper models for routine steps; the structural fix is moving Cline to a flat-rate endpoint so heavy days stop scaling the bill.
The single biggest Cline cost lever is file selection: one 5,000-line file mentioned in a 40-step task can account for the majority of the entire task's tokens.
Cline supports any OpenAI-compatible provider: set the base URL to https://api.stdcmpt.com/v1, model standardcompute, and Cline runs on a flat monthly price (from $39/mo) with smart routing choosing budget models for routine steps and frontier models for hard ones — automatically doing the model-splitting trick, without a token meter. ClinePass ($9.99) is the cheaper flat option if open-weight models cover your work; Standard Compute is the flat option that keeps the frontier.
Copilot and Cursor subsidize and cap usage behind a subscription; Cline passes raw API pricing through. You pay exactly what the models cost — which is transparent, but unbounded. That's why heavy Cline users end up on flat-rate endpoints.
For planning and complex refactors, yes — keep a frontier model there. For mechanical Act steps (applying edits, running commands, summarizing output), budget models are usually indistinguishable. Routing per-step is the cost/quality sweet spot.