Code editor Zed shifts to token-based AI pricing, adds new AI models, and expands access
Zed, a popular code editor, is shifting its Zed AI service from prompt-based limits to a token-based pricing model. This major update aims to align pricing more closely with the actual cost of operating AI features, as the prior prompt-based system often obscured expenses and led to misaligned incentives.
As part of this transition, Zed Pro customers are receiving three months' advance notice before migration begins. Depending on individual AI usage, monthly bills could rise or fall. New users are subject to the token-based model starting immediately, while current users will migrate over the next three months.
In addition to the billing overhaul, Zed has broadened options for using AI in the editor. Users can now bring their own API keys, run local models such as Ollama, or connect to third-party agents including Gemini CLI and Claude Code through the Agent Client Protocol (ACP). Building on these changes, Zed is expanding its hosted AI offerings to include GPT-5 (along with mini and nano variants), Gemini 2.5 Pro, Gemini 2.5 Flash, along the existing Anthropic models.
For those who prefer not to use AI, Zed still allows complete disabling of AI features in the code editor.
- Code Editor
- Freemium • Open Source
Related news
External links
- Zed's Pricing Has Changed: LLM Usage Is Now Token-BasedZed • Official source
Comments
Well... I'm using VS Code with OpenAI's Codex extension and a $20/mo. ChatGPT subscription. I can use at least the equivalent of $5 every 5 hours, all month long. That's the potential worth of at least $672/mo of even better quality use, because Codex-5-high is scoring at the top of the SWE benchmark leaderboard (how good it's at solving code problems in real life).
So $5 of AI-credit for a $10/mo subscription for a subculture IDE is a retarded choice to make from a financial point of view.
From Zed announcement: "LLM bills have become our biggest expense, and more paying customers translates to more money lost." That's the sad reality of all AI companies, the more successful they are, the more money they're loosing. Some AI companies redirects most of this cost to OpenAI or Anthropic (that gets billions in investments few times in a year) but it's only a time bomb that will blow once theses investments will seriously slowing down (that have already happen since less companies are willing to invest). And building a business model on top of the shoulder of an unprofitable company is suicidal. Running models on user device is a solution, but who's gonna pay to make all the work by himself.