Changes to GitHub Copilot individual plans

Original Article ↗ Hacker News Discussion ↗

Pricing & Plan Changes

  • Main shock: removal of Claude Opus 4.5/4.6 from Copilot Pro, replacement with Opus 4.7 at a much higher multiplier (e.g., 3x → 7.5x “introductory,” with fear it may rise further).
  • Pro+ now costs 4× Pro while Opus multipliers and quotas make effective Opus usage ~2× (or more) the previous price.
  • Some subscribers prepaid annually expecting stable pricing; many view mid-cycle changes as a “rug pull.”
  • GitHub paused new signups; some interpret this as capacity and/or cost pressure.

Rate Limits & UX Friction

  • New daily/weekly/session limits plus 5‑hour “windows” frustrate users who prefer bursty, project-based work.
  • People complain they can’t reliably consume the full advertised monthly “premium requests” due to sub-quotas.
  • Opus 4.7 in Copilot CLI/agents reportedly stalls mid-task, loses context after compaction, or wipes history.

Model Quality & Choice

  • Several say Opus 4.5/4.6 followed instructions better and improved productivity compared to GPT 5.x or smaller models.
  • Others argue most tasks work fine with cheaper models (e.g., Haiku/Sonnet) and that always using the “Ferrari” is wasteful.
  • There’s debate whether model choice should be manual or automatically selected by the system; skepticism that LLMs know when they’re failing.

Alternatives & Migration

  • Many consider canceling Copilot and moving to:
    • Claude Pro / Claude Code (though availability of Claude Code in Pro appears to have changed multiple times; people report conflicting information).
    • Other IDE harnesses (Cursor, OpenCode, Claude Code in VS Code).
    • Direct API access (Anthropic/OpenAI/GLM/Qwen) or local models to avoid “AI middlemen.”
  • Some liked Copilot specifically because it was per-request rather than per-token, and integrated into VS Code.

Trust, Business Model & Industry Dynamics

  • Users see this as emblematic of VC-subsidized AI now colliding with real inference costs, especially for long-running agentic workflows.
  • Some praise the blog’s initial transparency about refunds; others note that certain refund language was later removed.
  • Broad concern: building workflows or businesses on third-party AI platforms is risky due to sudden pricing and feature changes.