GitHub Copilot Limits Tighten as Agents Spike Compute Costs
GitHub pauses individual Copilot signups, adds token limits per session/week, restricts top models to $39/mo Pro+, due to agentic workflows burning 10x more tokens than six months ago.
Agentic Workflows Drive Compute Overload
Coding agents in Copilot now run long, parallelized sessions that consume far more resources than original plans supported—up to 10x more tokens than heavy users burned six months ago. This forces tighter usage limits to maintain reliability, as more customers hit caps from expanded agent capabilities doing heavier work. Builders relying on Copilot for agentic coding should expect per-session and weekly token-based throttling, shifting from prior per-request billing that eroded margins on token-intensive runs.
Specific Plan Restrictions Hit Individuals Hard
Individual plans see paused signups, Claude Opus 4.7 gated behind $39/month Pro+, and older Opus models dropped entirely. Affected features span Copilot CLI, cloud agents, GitHub.com code review, and IDE integrations in VS Code, Zed, JetBrains—check https://github.com/features/copilot/plans for details. Previously unique per-request pricing (like Windsurf's now-abandoned credit system) gives way to token limits, protecting GitHub from high-compute outliers.
Ambiguous Scope Complicates Migration
Announcement fails to specify which of Microsoft's 75+ Copilot-branded products (15 named GitHub Copilot) are impacted, per Tey Bannerman's mapping. Developers must infer from plans page, risking surprises in production workflows. Pair with Claude Code's pricing wobble (reversed $100/month) to see pattern: agentic AI's real costs are forcing providers to rethink free-tier generosity.