OpenAI, Cursor cut coding costs

OpenAI and Cursor moved within weeks of each other to make coding-agent costs easier to price. OpenAI’s Codex rate card, updated on Tuesday, lays out credits per million input, cached input and output tokens across ChatGPT plans. Cursor said on Monday it launched Composer 2.5, a model trained for longer-running coding jobs at lower token cost. Together, the two moves give software teams more explicit numbers for budgeting coding agents and sustained automation work. ### When did OpenAI change how Codex is billed? OpenAI said in its Help Center that it changed Codex pricing on April 2, 2026 for Plus, Pro, ChatGPT Business and new Enterprise customers, replacing per-message pricing with API-style token billing. The company said it extended that change on April 23 to all existing ChatGPT Enterprise plans, including Edu, Health, Gov and ChatGPT for Teachers. The Help Center page says the new format prices Codex in credits per 1 million input tokens, cached input tokens and output tokens. OpenAI said the format “replaces average per-message estimates with a direct mapping between token usage and credits,” giving users a clearer view of how different workloads consume credits. ### What numbers did OpenAI put on the table? (help.openai.com) OpenAI’s published table lists GPT-5.5 at 125 credits per million input tokens, 12.50 credits for cached input tokens and 750 credits for output tokens. The same table lists GPT-5.3-Codex at 43.75 credits per million input tokens, 4.375 credits for cached input tokens and 350 credits for output tokens. (help.openai.com) OpenAI also said Codex costs about $100 to $200 per developer per month on average, while noting that actual usage varies with model choice, the number of running instances, automations and use of fast mode. The company said users can monitor remaining credit in the Codex settings usage panel. ### What did Cursor say Composer 2.5 is built for? Cursor said on May 18 that Composer 2.5 was trained specifically for long-running coding tasks, according to Indian Express, which cited the company’s blog post. (help.openai.com) The report said Cursor described the model as more reliable on complex instructions and said the update also included changes to communication style and effort calibration. Indian Express reported that Cursor positioned Composer 2.5 around sustained developer workflows rather than short, one-off prompts. The report said the company attributed the update to scaled training, more complex reinforcement-learning environments and new learning methods. (indianexpress.com) ### Why do these pricing disclosures matter to engineering managers? OpenAI’s rate card formalizes cost buckets that had previously been harder to estimate under per-message pricing. By breaking out input, cached input and output token charges by model, the page gives finance and engineering teams a way to compare code review, generation and automation workloads against plan credits. (indianexpress.com) Cursor’s launch adds a second data point for buyers weighing whether long-running coding agents can be deployed more broadly. Indian Express said Composer 2.5 was introduced at cheaper token cost, aimed at developer tasks that run across longer sessions. ### What happens next? (help.openai.com) OpenAI said some Enterprise customers remain on a legacy Codex rate card and should contact sales for details, indicating the migration is not yet fully uniform across all accounts. The company also said GPT-5.3-Codex-Spark may be available as a research preview and that final credit rates for that model have not been set. (indianexpress.com) Cursor’s next step is user adoption inside its editor and coding workflows after the May 18 release. OpenAI’s next visible milestone is continued use of the token-based Codex card across Plus, Pro, Business and Enterprise-family plans, as listed on its Help Center page. (help.openai.com)