GPT-5.5 Pro today? What’s Behind Spud Rumors, Codex Leak, and the Anticipation of OpenAI

On April 23, a wave of expectations rose around OpenAI again: social media, Reddit and forecast markets are discussing the possible release of the GPT-5.5 Pro, the next big model after GPT-5.4. At first glance, everything looks like a classic pre-release vibe: leaked interfaces, strange code names, disappearing screenshots, high Polymarket and talk of “today is the day.” But if you separate the facts from the noise, the picture is much more careful.

Officially, OpenAI now published the line GPT-5.4. At Help Center, the company describes three main ChatGPT modes: Instant on GPT-5.3 Instant, Thinking on GPT-5.4 Thinking, and Pro on GPT-5.4 Pro. In API documentation, OpenAI also recommends starting with gpt‐5.4 as the flagship model for complex reasoning and coding. Frontier models include GPT-5.4, GPT-5.4 Pro, GPT-5.4 mini/nano and older GPT-5/GPT-4.1, but not GPT-5.5.

This is important: As long as there is no model card, changelog, pricing page or OpenAI help article on GPT-5.5, you cannot talk about the release as a fait accompli. The official Codex Models page now also indicates gpt‐5.4, gpt‐5.4-mini, gpt‐5.3-codex and gpt‐5.3-codex-spark. GPT-5.5 is not displayed there.

However, the rumors did not arise from scratch. The main reason is the alleged leak in Codex** on April 22. According to media reports, Codex's internal model selection menu was briefly visible to a limited number of users and included "GPT-5.5", "oai-2.1" and several code or experimental builds. Some users claimed to have tested the models before the interface started returning “model not found” errors. But OpenAI GPT-5.5 has not been confirmed, there are no official benchmarks, and early impressions remain anecdotal.

The second line is codename Spud. OpenAI has completed pretraining of the new model under the code name, and Sam Altman allegedly wrote to employees that the company is waiting for a “very strong model” for several weeks. This is not the same as the public release of GPT-5.5, but it explains why the market believed in the upcoming launch: if pretraining is completed at the end of March, and the management says “a few weeks”, the mid-to-late April window looks logical.

Greg Brockman added fuel to the fire. According to the media, he called Spud not just an incremental update, but a new base model / new pre-train, which included about two years of research. He also described the expected leap through the idea of “big model smell”: the model should better understand the context, instructions, and intentions of the user, so that there is less need to chew everything. Brockman has not given public benchmarks, pricing, release date or final commercial name.

The most controversial rumor is “Crest Pro Alpha.” The name is often associated with GPT-5.5 Pro, but there is little evidence. Several options are possible: it could be an A/B checkpoint, a temporary test build, a routed variant of the GPT-5.4 Pro, or several different internal models that the community has combined into one myth. This is a typical dynamic for AI releases: one screenshot from a model picker, some impressive demos, then YouTube, X and Reddit turn it all into a “new model found before release.”.

Separately discussed demos: Three.js scenes, Minecraft-like mini-games, SVG, voxel art, interfaces and 3D reconstructions. Spud/GPT‐5.5 Pro is being talked about leaps in 3D simulations, web development, SVG generation and creative game demos. But the problem is that it's a *ceiling demos, not a systematic assessment of reliability. One successful “one-shot” video doesn’t tell you how the model works on a 10th try, on a long repository, on a 40-step agent workflow, or on independent sets like SWE-bench and OSWorld.

Why April 23rd? Here, almost all the energy comes from prediction markets. Polymarket at the time of the audit showed “April 23” as the leading outcome in the market “GPT-5.5 released on...?” – about 93%. At the same time, market rules require public availability of the model and reliance on official OpenAI information. This means that the market itself recognizes that the final arbiter is not a rumor, but an official release.

On Reddit, the situation looks like a typical pre-release panic: users are discussing the flow of posts on X, possible links to outages, Polymarket and accesses through Bedrock/Foundry. But there is skepticism in the comments: some write “No”, others suggest Thursday, others remind that past models GPT-5 often released on Thursday. This is a useful mood barometer, but not a source of confirmation.

Now the most honest wording is this: OpenAI is probably preparing a large next model codenamed Spud; GPT-5.5 Pro may be its public name, but this is not proven. Codex leak has increased expectations, Polymarket almost “put” on April 23, and the media is already putting together a puzzle of pretraining, Altman/Brockman words and hidden tests. But official documentation for April 23 still points to GPT-5.4 as the current flagship lineup.

If the GPT-5.5 Pro does come out today, it’s not just a “slightly smarter GPT-5.4,” but a focus on agency, coding, long tasks, multimodal/visual reasoning, and possibly stronger frontend/3D/creative code tasks. This conclusion follows from where GPT-5.4 was headed: OpenAI positioned it as a model for professional work, reasoning, coding, agentic workflows, computer use, use tool and long context. Consequently, GPT-5.5, if it is a direct successor, will almost certainly be marketed as the next step in precisely these directions.

But for now, it's best to keep three scenarios in mind.

One: release today and the rumors will be an early reflection of the real rollout.

Second: release by the end of April because Spud is really in the final stages, but OpenAI is still waiting for infrastructure readiness.

Third, ** there will be no GPT-5.5 under this name** - Spud can come out as GPT-6, as GPT-5.4 Pro refresh, as Codex-only preview or as a set of specialized models.

GPT-5.5 Pro is expected and looks plausible, but not officially confirmed. “Spud” is the strongest signal, Codex leak is the main hype trigger, Polymarket is the expectation indicator, and GPT-5.4 Pro remains the last officially documented Pro-model OpenAI.