GPT‑5.5 pretraining reported
Reports say OpenAI has finished pretraining a model dubbed GPT‑5.5 and an April release is likely, though the company has not confirmed a launch. The same coverage notes work on a unified Codex coding app and a new “Scratchpad” feature that employees have hinted at publicly. (abhs.in) (testingcatalog.com)
A new report says OpenAI finished pretraining a model called GPT‑5.5 on March 24, but OpenAI had not announced a launch as of April 12. (abhs.in) The report, published April 11, said the model is internally codenamed “Spud” and pointed to a Polymarket contract pricing a release before April 30 at 78%. OpenAI has not published a GPT‑5.5 product page, system card, or release note. (abhs.in) (openai.com) Pretraining is the long first pass where a model learns patterns from massive datasets before later tuning and safety testing. OpenAI described GPT‑4.5 on February 27, 2025 as “a step forward in scaling up pre-training and post-training,” which is the same basic pipeline this new report is referring to. (openai.com) OpenAI’s recent public roadmap has centered on simplifying a crowded product lineup. Sam Altman wrote on February 12, 2025 that OpenAI wanted AI to “just work” and said the company wanted to move away from a complicated model picker. (community.openai.com) That cleanup is already visible in OpenAI’s current products. On March 5, 2026, OpenAI released GPT‑5.4 and said it combined recent advances in reasoning, coding, and agentic workflows into one model with up to 1 million tokens of context. (openai.com) The same consolidation push now extends to software. CNBC reported on March 19 that OpenAI plans to combine its ChatGPT app, browser, and Codex coding app into one desktop “super app,” with Applications chief Fidji Simo overseeing the effort and President Greg Brockman assisting. (cnbc.com) OpenAI has already laid the groundwork with Codex. The company’s Codex app page says the app runs Codex threads in parallel with built-in Git tools, automations, and worktree support, and OpenAI’s February 2 launch post called it “a command center for agents.” (developers.openai.com) (openai.com) TestingCatalog reported on April 11 that OpenAI is developing a “Scratchpad” inside Codex that would let users start multiple Codex chats from a to-do list and run them in parallel. TestingCatalog also said code references in the current client suggest broader tasks beyond software engineering and a heartbeat system for long-running jobs. (testingcatalog.com) Some of that direction matches OpenAI’s own documentation. OpenAI says Codex worktrees let multiple independent tasks run in the same project without conflicts, and its automations docs describe a sidebar “Triage” inbox for background runs that surface findings later. (developers.openai.com 1) (developers.openai.com 2) What is still missing is the part only OpenAI can confirm: the model name, benchmark gains, pricing, and release date. Until OpenAI posts a release note or product page, GPT‑5.5 remains a reported next step rather than an announced one. (openai.com)
