OpenAI retools subscriptions and models

OpenAI just split its paid ChatGPT plan in two directions instead of one: there is now a $100 a month Pro tier sitting below the existing $200 Pro tier, and both are built around how much Codex coding usage you need. Codex is OpenAI’s coding tool inside ChatGPT, and the new pricing is basically a meter for how long and how often you want that tool working. OpenAI says Plus at $20 is for lighter weekly use, Pro at $100 is for “real projects,” and Pro at $200 is for people running demanding workflows continuously. The new $100 plan gets 5 times the usage limits of Plus, while the $200 plan keeps 20 times the limits of Plus. For a temporary launch push that runs through May 31, 2026, OpenAI says the $100 plan gets 10 times Codex usage versus Plus. OpenAI also tied the new tier to a specific coding model called GPT-5.3-Codex-Spark, which it describes as a fast research-preview model for day-to-day coding tasks. That means the $100 plan is not just more messages; it is also a way to steer heavier coding users toward a faster specialized model. This pricing change lands after OpenAI spent March simplifying the model lineup inside ChatGPT. GPT-5.3 Instant became the default model for logged-in users, and OpenAI framed it as a smoother everyday model with better web answers and fewer unnecessary refusals. Above that default layer, OpenAI has been pushing GPT-5.4 Thinking and GPT-5.4 Pro as the heavier-duty options for hard tasks. OpenAI says GPT-5.4 folds in the coding gains of GPT-5.3-Codex while improving work across spreadsheets, documents, presentations, tools, and long-running agent workflows. The cleanup came with a lot of retirements. OpenAI says ChatGPT retired GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 Instant and Thinking on February 13, 2026, then retired GPT-5.1 Instant, Thinking, and Pro on March 11, 2026. The important split is between ChatGPT and the application programming interface, which is the developer pipe that lets outside software call OpenAI models directly. OpenAI says those retired models are gone from ChatGPT, but API access remains unchanged, so companies that built products on older models are not being cut off on the same schedule as consumer users. A separate OpenAI update on April 10 was about security, not pricing. The company said a GitHub Actions workflow in its macOS app-signing process downloaded a malicious version 1.14.1 of Axios on March 31, 2026, after the widely used third-party library was compromised in a broader software supply chain attack. App signing is the digital seal that tells your Mac a program really came from the company named on the label. OpenAI said the workflow had access to signing certificate material for ChatGPT Desktop, Codex, Codex command line interface, and Atlas, but it found no evidence that user data was accessed, systems or intellectual property were compromised, or software was altered. OpenAI is treating the certificate as compromised anyway, which is the cautious move when the seal itself might have been exposed. The company said it is revoking and rotating the certificate, requiring macOS users to update, and warning that older Mac app versions may stop working or stop receiving support after May 8, 2026. Put together, the week’s changes show OpenAI trying to do two things at once: make ChatGPT simpler on the surface by collapsing users onto newer GPT-5 models, and make the paid ladder more granular underneath by charging according to how much coding work people actually run. The security notice fits the same pattern of tightening the plumbing while the company pushes more users toward desktop apps, coding agents, and longer-running workflows.