Cursor Cost Calculator

Try it live →

Empirical tool for measuring Cursor token usage and cost. Manual calculator rates come from Cursor's official docs. If the docs do not disclose a current Cursor rate, the manual calculator does not model it. CSV replay can also use clearly labeled import-only estimates for retired historical labels that no longer exist in the current Cursor catalog.

What it does

• Budget mode — "I have $60/month, what do I get?" Finds the plan that fits your spend ceiling, then shows plan coverage, token yield, and estimated out-of-pocket spend.
• Token mode — "I use 10M tokens/month, what does that cost?" Finds the cheapest plan for manual token entry, with both a quick estimate and exact token-bucket entry, and separates estimated usage cost from plan-adjusted out-of-pocket spend.
• Cursor CSV replay — Import one exported monthly Cursor CSV, reuse the exact input/cache/output token columns, and replay that usage through the same plan recommendation math and cost-versus-coverage framing used by the token calculator.
• Weighted model mix — Split usage across models (60% Sonnet, 40% Opus). Per-model weights, normalized to 100%.
• Variant toggles — Max Mode, Fast, Thinking, Caching. Documented Max/long-context rates are modeled from the catalog rather than inferred.
• Caching — Anthropic models use cache_write + cache_read with re-read amortization. Everyone else uses cache_read only. Different systems, different math.
• Import replay controls — User API Key rows are included by default for a “Cursor only” estimate, with strict vs best-effort label mapping and approximate rows called out explicitly.
• Imported Composer handling — retired Composer rows are API-priced during replay, while current composer-2.5 usage stays in the included Auto + Composer pool.
• Monthly usage summary — Imported usage shows priced API tokens, cache-read share within priced API rows, non-cache priced tokens, approximate tokens, unsupported tokens, days used versus comparison days for the imported month or date span, and API tokens per used day.
• 38 models, 6 providers — Anthropic, OpenAI, Google, xAI, Cursor, Moonshot.

Matching an imported CSV in manual mode

• Quick estimate — Use API tokens priced from the imported summary as your manual token total, then copy Cache-read share (priced API) into the cache slider.
• Conservative baseline — Use Non-cache priced tokens as your manual total if you want to see a no-cache starting point.
• Closest replay match — Switch to Exact token buckets and copy the CSV-style buckets directly: Input with cache write, Input without cache write, Cache reads, and Output.

Running it

npm install
npm run dev

Tests:

npm test

Build:

npm run build

Tech

React 19, TypeScript, Vite, Tailwind CSS 4, Vitest.

Architecture

• App orchestration boundary: src/app/
  • calculatorState.ts defines reducer-owned source state and defaults
  • calculatorReducer.ts owns state transitions
  • calculatorSelectors.ts derives replay reports, recommendations, and presentation-ready result view models
  • recommendationPresentation.ts is the source of truth for result semantics shown in the UI
  • cursorImportPresentation.ts and cursorImportActions.ts own replay-summary presentation helpers and import-action wiring
  • useCalculatorController.ts coordinates UI wiring and import side-effect orchestration
• Current Cursor pricing catalog: src/data/cursor-pricing.json
• Current catalog accessors: src/domain/catalog/
• Model config defaults, reconciliation, and capability rules: src/domain/modelConfig/
• Recommendation math and plan comparison: src/domain/recommendation/
• Import replay catalog and CSV pipeline: src/domain/importReplay/
• Import-only historical replay models: src/data/importReplayHistoricalModels.ts
• Import replay label mappings and approximation rules: src/data/importReplayLabelMappings.ts
• Presentational UI components: src/components/
• Automated coverage: src/domain/*/__tests__/ and src/app/__tests__/

Not included

• Auto + Composer pool — Cursor doesn't disclose dollar amounts.
• Live usage tracking — There is no Cursor API integration here. Imported replay uses exported CSVs, not live account telemetry.
• First-party verified pricing for every retired import label — Best-effort import mappings are estimates and are labeled as approximate.
• Thinking token pricing — Not documented separately by Cursor.
• Teams/Enterprise — Different pricing structures.

License

MIT