OpenAI Updates Codex to Token-Based Pricing Model

OpenAI has introduced a new token-based pricing structure for Codex, replacing the previous per-message credit system for Business and new Enterprise users. While legacy plans like Plus and Pro currently retain their old rates, they will be migrated to the token-based system in the near future. This update allows for more granular tracking of costs based on input, output, and cached token usage.

Key Points

OpenAI is shifting Codex pricing from per-message estimates to a precise token-based model (credits per 1M tokens).
The new token-based rates apply immediately to ChatGPT Business and new Enterprise plans, while others use a legacy card during a transition period.
Credit costs vary significantly by token type, with output tokens being substantially more expensive than input or cached input tokens.
Fast mode usage doubles the credit consumption across all models and tasks.
The change is designed to provide clearer visibility into usage and align Codex costs with standard API token metering.

Sentiment

The overall sentiment is skeptical and somewhat negative. While some commenters acknowledge the inevitability and even the fairness of token-based pricing, the dominant tone is one of concern about rising costs, frustration with opaque pricing structures, and anxiety about what this means for the sustainability of AI-dependent workflows. Many commenters see this as confirming that AI has been a subsidized bubble, and there is considerable discussion about migrating to cheaper alternatives.

In Agreement

Token-based pricing is more transparent and accurate than per-message pricing, providing clearer visibility into how input, cached input, and output tokens affect costs
This change was inevitable as AI subscriptions were heavily subsidized and unsustainable at current pricing levels
Aligning pricing with actual token usage is necessary for the industry to mature and become financially sustainable
The change may actually benefit users who manage their context well, since cached tokens are cheaper
Moving toward usage-based pricing creates proper financial back-pressure that discourages wasteful token consumption

Opposed

Using credits as an intermediary currency is deliberately confusing and designed to obscure actual costs, similar to mobile game gem-pack tactics
This feels like a rug pull — changing pricing without adequate notice while simultaneously ending the 2x credit promotion makes it harder for solo developers and small projects to continue
The pricing change suggests AI providers are running out of runway rather than succeeding, undermining the AI hype narrative
Users cannot predict token costs ahead of time due to hidden reasoning tokens and agentic loops, making token-based billing problematic for budgeting
All major AI providers tightening limits simultaneously suggests either they are commodity sellers with no moat, or the technology has plateaued