GitHub Copilot Moves to Token-Based Billing

Microsoft is moving GitHub Copilot from a fixed-request subscription model to token-based consumption billing, with rollout slated for the beginning of June and an announcement expected April 23. Internal documents and multiple reporting outlets show the company has paused new sign-ups for Pro, Pro+, and Student tiers, tightened rate limits for individual accounts, and removed Claude Opus access from cheaper plans to rein in runaway compute costs. The weekly operating cost for Copilot reportedly nearly doubled since January, driving the change.

The new billing charges for tokens separate input and output consumption and ties cost to model choice and token burn. Example figures disclosed include GPT-5.4 priced at about $2.50 per million input tokens and $15 per million output tokens in some reports, with parallel euro-denominated figures around €2.30 and €14 in others. Users will continue to pay a base monthly subscription but will receive token-credit allotments instead of fixed request counts. Reported enterprise allotments include $19/user/month with $30 pooled credits and $39/user/month with $70 pooled credits; pooled credits allow organization-wide sharing of token budgets. The change is being accompanied by tighter per-account rate limits and model availability restrictions for lower-priced tiers.

Token-based billing makes per-call costs visible, but it also forces teams to manage token budgets and optimize prompts, model selection, and caching. Practitioners should instrument token usage per pipeline, add cost-aware guardrails in CI/CD, and consider model fallbacks or smaller context windows for high-volume tasks. Expect these practical actions:

• •Implement tokens accounting and cost dashboards in observability tooling.
• •Add rate-limit backoffs, batching, and response truncation to reduce output token burn.
• •Re-evaluate model selection and use GPT-5.4 or cheaper variants for lower-value tasks.

This move follows a broader industry trend where providers shift from flat-rate subscriptions to consumption billing as model capabilities and per-call costs rise. Anthropic and other vendors recently tightened limits or moved enterprise customers to usage-based pricing. For Microsoft, the change reduces subsidy of heavy usage and aligns Copilot with cloud-native, consumption-based business models. For the developer economy, the timing matters: widespread adoption of LLM-powered coding assistants increased token burn substantially, and companies must now reconcile productivity gains with predictable budgets.

Track the official pricing announcement for final token rates, how Microsoft handles legacy subscribers, and whether reduced access to models like Claude Opus persists. Practitioners should prepare by instrumenting token metrics, rehearsing budget alerts, and updating procurement policies for AI compute spend.

The shift moves cost transparency into the hands of users and teams, but it also adds operational overhead: token-metering, prompt engineering, and cost governance become first-class concerns for any organization embedding Copilot into engineering workflows. Expect short-term friction as companies adapt and long-term normalization toward consumption-based AI tooling.