Max plan: fast mode incorrectly billing to extra usage at low plan utilization

Bug

On a Claude Code Max plan, enabling fast mode (/fast) caused usage to be billed to extra usage credits instead of the plan quota, despite very low plan utilization.

Evidence

• Plan: Claude Code Max
• Weekly usage: 15% (All models), 2% (Sonnet only)
• Session usage: 2%
• Extra usage charged: $23.84 — exceeding the $20 monthly spend limit
• Trigger: Enabled fast mode via /fast, used ~3 conversation turns, then received a message that fast mode was disabled due to overusage

Expected behavior

At 15% weekly plan utilization, fast mode (same Opus 4.6 model, faster output) should be covered by the Max plan quota — not routed to pay-per-token extra usage.

Actual behavior

Fast mode tokens were charged to extra usage credits, burning through ~$24 in a few turns despite the plan being far from its limits.

Request

• Investigate why fast mode usage was billed to extra usage instead of plan quota
• Refund the extra usage charges that should have been covered by the Max plan

Environment

• Platform: Windows 11 Pro
• Model: claude-opus-4-6 (1M context)
• Claude Code: CLI