OpenAI carves a paid Codex tier

OpenAI just inserted a new price step between a $20 plan and a $200 plan because coding agents are expensive enough that one flat subscription no longer fits everyone. Its Codex pricing page now shows a Pro option starting at $100 a month, with access to GPT‑5.3‑Codex‑Spark and higher request limits than ChatGPT Plus. (openai.com) That is a shift from “pay one fee and use the chatbot” to “pick a lane based on how much machine time your coding work burns.” OpenAI’s help page says the $200 Pro plan is still the highest-usage tier and now offers 20 times the Codex usage of Plus, while the $100 plan is a lower-usage Pro option. (help.openai.com) Codex is OpenAI’s coding product, and it is not just autocomplete in a text box. OpenAI describes it as an agentic coding system that can plan work, build features, refactor code, review changes, and run parallel tasks across projects in cloud environments. (openai.com) The model behind the new tier is GPT‑5.3‑Codex‑Spark, which OpenAI labels a research preview tuned for near-instant coding iteration. OpenAI’s model page says that version is available to ChatGPT Pro subscribers, while heavier or deeper coding work can use other Codex models. (openai.com) OpenAI had already signaled that this product would not stay cheap for power users. Its Codex rate card says average spending runs about $100 to $200 per developer per month, with costs varying by model, number of running instances, automations, and use of fast mode. (help.openai.com) That cost structure helps explain why OpenAI is separating ordinary chat use from developer use. The Economic Times reported that the new $100 plan gives five times the Codex usage of the $20 Plus tier and is aimed at longer, more intensive coding sessions. (economictimes.indiatimes.com) The same report said OpenAI is also moving third-party integrations such as OpenClaw out of standard subscription buckets and into pay-as-you-go billing. That means outside tools that call Codex start looking less like a buffet and more like a taxi meter. (economictimes.indiatimes.com) The competitive target is easy to spot because Anthropic already sells a premium Claude Max tier from $100 a month. Anthropic’s pricing page says Claude Max offers either 5 times or 20 times more usage than Claude Pro, plus priority access and higher output limits. (anthropic.com) So the fight is no longer just “whose model writes better code.” It is becoming “whose pricing lets a software team hand more real work to an agent without getting surprised by limits or bills,” and OpenAI is now carving Codex into the same premium band where Anthropic has been selling Claude Max. (openai.com) (anthropic.com) OpenAI also has a technical reason to charge more here. Its GPT‑5.3‑Codex system card says the model combines the coding strength of GPT‑5.2‑Codex with the reasoning and professional knowledge of GPT‑5.2 so it can handle long-running tasks that involve research, tool use, and complex execution. (openai.com) That kind of model behaves less like a spellchecker and more like a junior engineer who opens files, runs tools, and keeps working for a while. Once products start doing that, companies stop pricing them like chat windows and start pricing them like labor-saving infrastructure. (openai.com) (help.openai.com)