Cursor Frontier Models Move to Token-Based Max Mode Billing on March 16

Releasebot·high signal

Starting today, GPT-5.4, GPT-5.3 Codex, Claude Opus 4.6, Claude Sonnet 4.6, and Claude Opus 4.5 in Cursor switch from request-based to token-based billing under Max Mode, meaning a single complex prompt can consume 10–50x more credits than before. This is a significant cost change for heavy users who previously benefited from flat request counts on frontier models. Community recommendation is to route 80% of daily work to cost-efficient models (DeepSeek V4 Lite, Gemini 3 Flash) and reserve Max Mode for genuinely complex refactors.