Zed Switches to Token-Based AI Billing, Cuts Pro to $10, Adds GPT‑5 and Gemini

Zed is replacing prompt caps with token-based billing, lowering Pro from $20 to $10/month and including $5 in credits, with extra usage at list price +10%. Hosted access adds GPT-5 (mini/nano) and Gemini 2.5 Pro/Flash alongside Claude Sonnet/Opus; the Free plan keeps 2,000 accepted edit predictions but drops hosted prompts. Migration runs through late 2025 and users can bring their own keys, use local models, or disable AI entirely.

Key Points

• Pricing switches from prompt-based to token-based with usage billed at API list price +10%, and Pro now includes $5 in credits.
• Pro price drops 50% to $10/month, and hosted access expands to GPT-5 (+ mini/nano) and Gemini 2.5 Pro/Flash alongside Claude Sonnet/Opus.
• Free plan keeps 2,000 accepted edit predictions but removes the monthly Zed-hosted prompt allowance.
• Users can BYO API keys, use local models (Ollama), connect third-party agents via ACP, pay via hubs (Copilot, OpenRouter, AWS Bedrock), or disable AI.
• Rationale: align costs with value, reduce complexity, and sustain investment in the editor; consistent policy of list price +10% for new hosted models.

Sentiment

Overall, the sentiment of the discussion is **mixed to predominantly negative**. While some users understand the financial necessity for Zed to adopt token-based pricing, a significant portion expresses frustration with the unpredictability and perceived obfuscation of this billing model. Furthermore, many comments reveal disappointment with Zed's core editor development and the quality of its AI features compared to competitors, suggesting a mismatch between user expectations and Zed's current product focus. The broader discussion also reflects skepticism about the sustainability of third-party AI tool providers in the long run.

In Agreement

• Token-based pricing generally makes sense for companies like Zed to align costs with usage.
• The pricing change is seen as fair and necessary to make the LLM integration financially viable for Zed.
• Prompt-based billing was seen as less transparent and obscured true costs, making token-based pricing a more 'honest' approach.
• For users who value Zed primarily as an editor and use AI sparingly or with their own keys, the new pricing with a lower base subscription could save money and better reflect their usage.
• The 'bring your own key' (BYOK) model is appreciated by corporate purchasers for governance, budget controls, and flexibility to use different frontends.

Opposed

• Token-based pricing leads to unpredictable and difficult-to-forecast spending for users and teams.
• It creates misaligned incentives where AI providers (and potentially integrators) benefit from more verbose inputs and outputs, which may not always equate to better value for the user.
• Many users feel Zed's editor development has stagnated or shifted too much towards AI features since their introduction, neglecting core editor performance (e.g., handling large files, memory usage) and extension support.
• Zed's AI features, particularly edit prediction, are perceived as significantly inferior to competitors like Cursor, diminishing the value of the AI integration.
• The business model of third-party AI tooling companies (like Zed or Cursor) is seen as precarious and unsustainable, as direct offerings from LLM providers (e.g., Claude Code, OpenAI Codex) increasingly offer 'unlimited' usage plans and superior integrations, making intermediary services less appealing.
• Tokens are considered an 'implementation detail' and a form of 'deliberate obfuscation' that complicates user understanding of costs and value.
• The value proposition of Zed Pro is questioned, with some users seeing little reason to subscribe if they can bring their own API keys.