OpenAI updates pricing tiers
OpenAI refreshed its API and Codex rate cards, listing separate tiers for GPT-5.4, GPT-5.2 and GPT-5.1 and adding cached-input and Batch API pricing that change unit costs by use case. The published rate cards and a companion Codex credit schedule show how different plans (Plus through Enterprise/Edu) are being routed and charged. ( )
OpenAI has redrawn how developers pay for its newest models, with separate rate cards for GPT-5.4, GPT-5.2 and GPT-5.1 and new discounts for cached prompts and batch jobs. (openai.com) On OpenAI’s live API pricing page, GPT-5.4 is listed at $2.50 per 1 million input tokens, $0.25 for cached input tokens, and $15 for output tokens. GPT-5.4 mini is $0.75 in and $4.50 out, while GPT-5.4 nano is $0.20 in and $1.25 out. (openai.com) OpenAI says the Batch Application Programming Interface cuts input and output prices by 50% for jobs that can run asynchronously over 24 hours. The same page says regional processing for data residency adds 10%, and the standard GPT-5.4 rates apply to context lengths under 270,000 tokens. (openai.com) Older GPT rows are still in the published pricing stack. OpenAI’s GPT-5.2 model page lists GPT-5.2 at $1.75 per 1 million input tokens, $0.175 for cached input, and $14 for output, and BenchLM’s April 13, 2026 pricing roundup says OpenAI still publishes GPT-5.1 at $1.25 in, $0.125 cached input, and $10 out. (developers.openai.com, benchlm.ai) The key pricing change is not just the headline token rate. OpenAI now publishes cached-input prices separately, which means repeated prompt prefixes such as a fixed system prompt or long instructions can be billed at one-tenth of the normal input rate on the GPT-5.4 family. (openai.com, benchlm.ai) That shifts the economics for companies running agents, coding tools, and other applications that send the same setup text over and over. A workload that can use prompt caching and the Batch Application Programming Interface will land on a different effective price than one that pays the standard rate on every call. (openai.com, benchlm.ai) OpenAI also updated how Codex is charged inside ChatGPT plans. Its Help Center says that, as of April 2, 2026, Codex pricing for new and existing Plus, Pro, and Business customers, plus new Enterprise customers, moved from per-message pricing to token-based pricing aligned with the Application Programming Interface. (help.openai.com) The new Codex rate card lists GPT-5.4 at 62.50 credits per 1 million input tokens, 6.250 credits for cached input, and 375 credits for output. The same table lists GPT-5.2 and GPT-5.3-Codex at 43.75 input credits and 350 output credits, and says Fast mode uses twice as many credits. (help.openai.com) OpenAI says existing Enterprise and Education customers, along with new and existing Education, Teacher, and Healthcare plans, should stay on the legacy Codex rate card until migration happens in the coming weeks. The Help Center article was updated 19 hours before it was crawled on April 14, 2026. (help.openai.com) The result is a pricing menu that now depends as much on workflow design as on model choice. For developers comparing GPT-5.4 with GPT-5.2 or GPT-5.1, the cheapest published row is no longer always the cheapest bill. (openai.com, developers.openai.com, benchlm.ai, help.openai.com)
