GPT‑5 launch whispers
Rumors are heating that OpenAI could announce a GPT‑5‑branded release soon, with one roundup saying prediction markets put a 78% chance of a launch by April 30 (findskill.ai). At the same time, competitors and model updates are tightening the field—FreeCodeCamp notes GLM‑5.1 improved benchmarks and an industry digest says GPT‑5.4 recently beat humans on a desktop‑productivity test ( ).
OpenAI has already launched a model called GPT‑5, and the current whisper campaign is really about whether another GPT‑5‑branded release lands before April 30. (openai.com) OpenAI’s GPT‑5 page says the model is “available to everyone” in ChatGPT, while the company’s research index shows GPT‑5.4 was released on March 5, 2026. A separate OpenAI product page says GPT‑5.4 added native computer use and a 1 million token context window. (openai.com; openai.com; openai.com) That timing matters because some rumor roundups are pointing to prediction markets instead of company announcements. Manifold’s April 2026 model-release market showed an 81% chance on April 14 for “GPT‑5.5 (or 5.x variant)” to arrive during April, but that is a bet on a future event, not a confirmation from OpenAI. (manifold.markets) For readers who have not tracked the last few months, the naming has gotten messy. OpenAI’s public lineup now spans GPT‑5, GPT‑5.4, GPT‑5.4 Pro, GPT‑5.4 mini, and GPT‑5.4 nano, so a “GPT‑5 launch” rumor can mean a new flagship, a point upgrade, or a broader ChatGPT rollout. (openai.com; openai.com; openai.com) The competitive pressure is real even if the exact next OpenAI release date is not. FreeCodeCamp reported on April 13 that Zhipu AI’s GLM‑5.1, updated on March 27, posted a 28% improvement over GLM‑5 and reached 50 on the Artificial Analysis Intelligence Index, while also claiming 94.6% of Claude Opus 4.6 on coding benchmarks. (freecodecamp.org) OpenAI is still pushing the frontier on some tasks. Its March 5 release post said GPT‑5.4 was the first general-purpose model the company shipped with native computer-use capabilities, and an OpenAI community post said it scored 75.0% on OSWorld‑Verified. (openai.com; community.openai.com) That OSWorld‑Verified test measures whether a model can use software the way a person does, with screenshots, mouse clicks, and keyboard actions. Asanify’s April 14 digest said GPT‑5.4’s 75% score topped a 72.4% human baseline on desktop-productivity tasks inside apps such as Slack, Excel, and web browsers. (asanify.com) The hardware and cost gap has not disappeared just because open-weight models are closer on paper. FreeCodeCamp said self-hosting GLM‑5 requires about 1,490 gigabytes of memory, while OpenAI’s developer documentation lists GPT‑5.4 and GPT‑5.4 Pro with a 1.05 million token context window through the application programming interface. (freecodecamp.org; developers.openai.com) So the cleanest read on April 14 is narrower than the rumor mill suggests. GPT‑5 is not a pending debut anymore; the unresolved question is whether OpenAI adds another GPT‑5.x release before the month ends, while rivals keep shrinking the performance gap. (openai.com; manifold.markets; freecodecamp.org)
