Real-world stress test: Codex Pro cost ≈ $152 for 5 hours of heavy use

OpenAI’s Codex pricing is now being tested in public by users trying to map subscription tiers to actual heavy-use workloads. A May 19 post on X by user hqmank said sustained, high-volume Codex use worked out to about $152 for five hours of run time, adding a concrete field estimate to a pricing system OpenAI shifted to token-based accounting on April 2. OpenAI’s current pricing pages show Codex is bundled into ChatGPT plans including Plus and Pro, with higher limits on Pro and separate pay-as-you-go options for some business and API use cases. The result is a pricing debate centered less on list price alone than on how fast credits and limits are consumed in practice. ### How is Codex priced now? OpenAI said on its help page that it moved Codex to token-based pricing on April 2, 2026, replacing per-message estimates with a system tied to input, cached input and output tokens. The company said the change applied to Plus, Pro, Business and Enterprise plans, with Enterprise migration completed on April 23 for existing customers including Edu, Health and Gov. (help.openai.com) The Codex rate card lists usage in credits per 1 million tokens by model. For GPT-5.3-Codex, OpenAI lists 43.75 credits for input tokens, 4.375 for cached input tokens and 350 for output tokens; for GPT-5.5, it lists 125, 12.50 and 750 credits respectively. OpenAI says actual credit usage depends on the mix of inputs, cached inputs, outputs and whether fast mode is used. (help.openai.com) ### What do the Pro tiers actually include? OpenAI’s Codex pricing page says Pro starts at $100 a month and offers either 5x or 20x more usage than Plus, depending on tier. The same page says the $100 Pro tier is temporarily doubled until May 31, 2026, and shows Pro as including access to GPT-5.3-Codex-Spark in research preview. OpenAI’s Pro help page says the $100 plan is aimed at users doing “real projects” and normally includes 5x higher limits than Plus, while the $200 plan includes 20x higher limits. (help.openai.com) The help page also says the current promotion gives the $100 plan 10x Codex usage versus Plus for a limited time. ### Where does the “$152 for five hours” figure come from? (developers.openai.com) A May 19 X post cited in the source briefing attributed that estimate to user hqmank, who described a heavy-use stress test and criticized what the post called token hoarding. That figure does not appear in OpenAI’s official pricing pages, which describe average monthly cost ranges rather than a fixed five-hour burn rate. The user estimate is therefore a real-world anecdote, not an OpenAI benchmark. (help.openai.com) OpenAI says on its Codex rate-card page that “on average” Codex costs about $100 to $200 per developer per month, while adding that variance is large depending on model choice, number of running instances, automations and use of fast mode. That company estimate is the closest official reference point against which user stress tests can be compared. (help.openai.com) ### Why are users focusing on limits instead of just subscription price? OpenAI’s April 9 community post introducing the new $100 Pro tier said the plan offered 5x more Codex usage than Plus and was intended for longer, high-effort sessions. That announcement also said the launch promotion would raise the $100 plan to as much as 10x Plus usage through May 31. (help.openai.com) A separate OpenAI community reply on April 11 showed at least one user complaining that they hit a five-hour limit after about 1.5 hours of personal-project coding and “just 3 prompts.” That complaint, posted in OpenAI’s own forum, illustrates why users are comparing published multipliers with practical throughput under different workloads. (community.openai.com) ### What should users watch next? May 31, 2026, is the next concrete date in OpenAI’s current Codex pricing schedule. OpenAI says the temporary doubled Codex usage on the $100 Pro tier runs through that date, after which the standard 5x-versus-Plus framing is scheduled to matter more directly for subscribers comparing tiers. (developers.openai.com) (community.openai.com)