Max plan (/mo) runs out of quota before completing single feature — GPT-4.5 at finishes same task

What happened

Working on a multi-tenant database migration for a Nim backend project (12 tasks in a plan). Claude Code with Opus on the Max plan ($100/mo) hit the daily rate limit partway through Task 3 of 12.

The first session's context window bloated to the point where it couldn't be resumed, forcing a fresh session. The fresh session then burned through the remaining daily quota before finishing.

The same full task set was completed by ChatGPT 4.5 on a $20/mo plan without hitting any limits.

Details

• Plan: Max ($100/mo)
• Model: Opus (claude-opus-4-6)
• Workflow: Subagent-driven development — orchestrator dispatches Opus agents per task, Sonnet agents for spec review
• Tasks completed before limit: ~2.5 out of 12
• Rate limit message: "You've hit your limit · resets 7pm (Atlantic/Canary)"
• Context bloat: First session context grew too large to resume, requiring a fresh start that consumed additional quota re-reading the codebase

The math

• Claude Max: $100/mo, couldn't finish 1 feature in 1 day
• ChatGPT Plus: $20/mo, finished all 12 tasks same day

Suggestions

• Agentic workflows with subagents are the recommended pattern but they burn quota extremely fast since each agent gets a full context window. The quota needs to account for this usage pattern.
• Context window bloat forcing a fresh session effectively doubles the cost of a task. Better context management or session compression would help.
• At 5x the price, users expect at least parity with competitor output capacity.

Environment

• Claude Code CLI v2.1.81
• macOS
• Nim backend project with ~15 model files, ~600 line repository macro