Cursor removed cost information from the usage page and CSV export

Original Article ↗ Hacker News Discussion ↗

Cursor’s recent move to hide detailed dollar costs and token usage from its UI and CSV exports is drawing strong criticism from paying users, who see it as a user-hostile step that obscures how much they spend on AI coding assistance. Many argue that with competitive alternatives like Claude Code, Codex, Pi, Zed, and plain VS Code or Neovim harnesses, Cursor’s value proposition is eroding—especially as its pricing, UI changes, and Musk-era ownership raise concerns about trust, transparency, and long‑term reliability. Others point to wildly different token efficiencies across agent “harnesses” as a reason why clear cost metrics are essential for teams trying to manage budgets and choose the right tools.

Pricing opacity and cost visibility

  • Many see removal of dollar costs from the usage page, CSV export, and in‑IDE indicators as user‑hostile and designed to hide real expenses.
  • Several note it’s now easier to accidentally burn through included credits or leave expensive models enabled.
  • A Cursor employee says the CSV breakage was an accidental bug and is fixed, and that an in‑app dollar graph was removed because people misread it as actual spend, not API‑equivalent value.
  • Others dispute this, saying the Spending page doesn’t show enough detail for non‑admins and that users can understand the difference between subscription value and metered charges.

Perceived value of Cursor vs alternatives

  • Some long‑time users say they’ve largely abandoned Cursor for Claude Code, Codex, Zed, VS Code, Neovim, or custom CLIs, citing better agents, lower cost, or lighter tooling.
  • Cursor’s advantages mentioned:
    • Integrated IDE + agent experience (especially for deep code review and remote dev).
    • Strong autocomplete (“Cursor Tab”) and good agent UI.
    • Easy switching among many models (Anthropic, OpenAI, xAI, etc.).
    • Less noisy permissions model and workspace‑scoped run modes.
  • Downsides:
    • As a VS Code fork, it’s buggier and missing some extensions/features.
    • Recent UI shifts push “agent mode” over classic IDE usage and change keybindings, tabs, and entrypoints in ways some find disruptive.
    • Several feel plain VS Code plus plugins/CLI agents now beats Cursor.

Token usage, harness design, and cost efficiency

  • One contributor shares comparative token usage across different agent harnesses (Codex, Claude Code, Pi, OpenCode, etc.) on the same OpenAI model, showing order‑of‑magnitude differences.
  • Discussion highlights:
    • Many popular harnesses heavily “spam” the context window with tools, memory systems, and verbose prompts, often inflating cached and uncached tokens.
    • Minimalist harnesses can complete typical coding/ops tasks with far fewer tokens, sometimes faster, but may lack orchestration features for long/complex sessions.
    • Others caution that more tokens can be justified by better reliability, long‑running workflows, and heavy use of caching.
    • Strong consensus that context window management, caching, and prompt minimalism materially affect costs.

Trust, ownership, and ethics

  • Several commenters link the changes to Cursor’s new ownership, expecting more aggressive monetization and “enshittification.”
  • Some express distrust about IP handling and vow to reduce or end corporate use.
  • There is heated debate over the branding of Grok and broader political/ethical concerns, with some dismissing them and others saying they’re disqualifying.