Zed's Pricing Has Changed: LLM Usage Is Now Token-Based

Original Article ↗ Hacker News Discussion ↗

Reaction to Zed’s New Pricing

  • Many say per-token pricing was inevitable and more honest than “unlimited” tiers that later get tightened; some still perceive it as a bait‑and‑switch given how soon it followed agentic editing.
  • Several users now see little reason to pay for Zed Pro versus bringing their own API key, especially given the 10% markup over provider list prices. Others are happy to pay $10/month to support the editor and get edit prediction plus $5 of tokens.
  • Some want additional tiers: “BYOK only”, or “edit prediction only” with no hosted LLM spend.
  • Zed staff emphasize that LLM resale is not the core business, Pro is optional, and you can set spend limits (including $0) so you can’t run up extra charges unintentionally.

Token-Based Pricing and Cost Predictability

  • Developers report widely varying real-world costs: from a few dollars per day per engineer to thousands per month for an org; horror stories of $500/day in other tools also appear.
  • Many find tokens extremely hard to reason about and forecast, especially compared to fixed message quotas or rate limits. There’s interest in better AI FinOps and usage analytics.
  • Concerns: “house always wins” credit systems, vendors tweaking tokenization or verbosity, incentives to stuff prompts or outputs.
  • Counterpoints: competition among model vendors and option to self-host big models at scale somewhat cap abuse; tokens are no more opaque than other infra units like GB‑seconds.

Incentives, Business Models, and First‑Party Tools

  • Some criticize “AI intermediaries” whose entire model is marking up OpenAI/Anthropic, calling it fragile and misaligned; others argue Zed adds real value via context management and UI.
  • Fear: editors and SaaS tools will gate basic operations behind AI to monetize every action.
  • Several predict first‑party agents (Claude Code, Codex, Gemini, etc.) plus CLIs/ACP-style protocols will dominate, leaving little room for multi-provider tools like Cursor/Windsurf.

Edit Prediction and Competition

  • Multiple comments say Zed’s edit prediction is far behind Cursor (and sometimes Windsurf / Copilot / JetBrains), though still occasionally valuable.
  • Some users pay Zed purely for predictions while using Claude Code or other tools for heavy lifting; Zed says a major investment in prediction quality is underway and model weights are open.

Zed as an Editor vs. AI Platform

  • A camp values Zed mainly as a fast, pleasant editor and collaborative environment, using LLMs only as “glorified Stack Overflow.”
  • Others feel core editor work has stagnated since AI arrived: issues with very large files, project sizes, macOS/Linux font rendering, and missing ecosystem features/extensions.
  • There’s recurring anxiety about VC funding (Sequoia) leading to long‑term “enshittification,” contrasted with admiration for Zed’s technical quality, ACP work, and openness about pricing.