[TechCrunch] GitHub Copilot Switches to Token-Based Billing, Devs Report Massive Cost Increases

GitHub is overhauling its Copilot pricing model, moving from a flat monthly subscription to a **token-based billing system** that has sparked immediate backlash across developer communities. The changes take effect **1 June 2026**.

What changed

Copilot previously charged a flat monthly rate per seat. Under the new model, developers are billed based on the number of tokens consumed during each session. Tokens count both input (prompts, context, code) and generated output. Some users report costs running **significantly higher** than the old flat rate.

One Reddit user described the increase as "holy f--- how much money was Copilot losing," with another posting that pricing "is no longer cost-effective or useful in any practical way." Screenshots circulating on social media show dramatically inflated bills compared to the previous subscription tier.

Why it matters for self-hosted AI

This move highlights a growing tension in the AI tooling ecosystem. Proprietary, cloud-dependent coding assistants like Copilot now carry **unpredictable usage costs** that scale with how much you actually use them. For the self-hosted AI community, this strengthens the case for local alternatives:

• Local coding agents like Claude Code, Codex, and Kimi Code run on your own hardware with no per-token charges
• Open-source models (Llama 3, Qwen, DeepSeek) integrated via Ollama or vLLM give you predictable infrastructure costs
• Tools like Open WebUI and n8n AI agents let you build AI workflows without vendor lock-in

Some defenders of the new pricing argue that heavy token consumption is driven by "vibe-coding" — developers who iterate endlessly without precise prompting. But critics counter that Microsoft's own tooling encouraged this behaviour by making it easy to burn through massive token volumes on premium requests.

The bigger picture

The shift to metered AI pricing isn't unique to GitHub. OpenAI, Anthropic, and other providers all charge by token for API access. What's new here is applying that same model to an **integrated IDE tool** that was previously a simple subscription. This could accelerate the migration toward locally-hosted coding agents where costs are fixed and predictable.

For developers already running local models, see our guides on setting up coding agents with local models and comparing Claude Code vs Codex vs Kimi Code.

**Source:** TechCrunch — 'What a joke': GitHub Copilot's new token-based billing spurs consternation among devs