OpenAI announced a major shift in Codex pricing, replacing average per-message estimates with token-based API billing for all users. Credits are now calculated based on input, cached input, and output tokens, with GPT-5.3-Codex charging $100-$200/month per developer on average. This change aims to provide clearer visibility into how token consumption impacts credit usage across tasks.

Legacy plans (Plus/Pro/Edu) will retain approximate credit-per-message rates until migration, while Business/Enterprise customers must adopt the new system. The update aligns pricing with actual token metrics, though fast mode doubles credit costs. Developers can monitor usage via Codex’s Usage panel, where token breakdowns reveal why some tasks consume more credits than others.

The FAQ clarifies that output-heavy operations and fast mode significantly increase expenses. Enterprises will receive migration timelines via email, while smaller teams should watch release notes for transition dates. OpenAI supports both rate cards temporarily, emphasizing flexibility during the shift.

This move reflects broader industry trends toward granular usage tracking. Developers must now optimize token efficiency to manage costs, particularly for automated workflows. OpenAI’s documentation stresses that final credit impacts depend on workload mix, with unpredictable variances possible between legacy and new systems.