GPT Image 2: Rumours, Leaks & Release Date (2026)

There is no official GPT Image 2 yet. As of April 2026, OpenAI has not published a model page, API alias, blog post, or announcement for anything called gpt-image-2 or GPT Image 2. But the rumours are specific enough (and the evidence coherent enough) that it's worth laying out exactly what's confirmed, what's plausible, and what's still pure speculation. Here's everything we know about GPT-Image-2, the leaks driving the chatter, and a realistic read on when it might arrive.

TL;DR

• OpenAI has not announced GPT Image 2 — the current public lineup stops at gpt-image-1.5 (released December 2025)
• Anonymous image models codenamed "packingtape," "maskingtape," and "gaffertape" reportedly appeared on LM Arena in early April 2026, with community testers praising their text rendering and world knowledge
• OpenAI has not confirmed the tape models are theirs, and the final public name may not be "GPT Image 2" at all
• No release date is confirmed; if Arena testing is underway, a launch within weeks or months is plausible — but that's inference, not a forecast.

What Is GPT Image 2?

GPT Image 2 refers to the anticipated next major version of OpenAI's natively multimodal image generation model family. The current public lineup is gpt-image-1.5, chatgpt-image-latest, gpt-image-1, and gpt-image-1-mini, alongside the deprecated DALL·E 2 and 3. No model using the identifier gpt-image-2 currently exists in OpenAI's public model documentation, API, or "All models" page.

The GPT Image family (introduced with gpt-image-1) is designed to use broad world knowledge when generating images, not just the text in your prompt. That's a meaningful architectural distinction from earlier Text to Image models, and it's exactly where the rumoured improvements for a potential GPT Image 2 model are expected to show up.

GPT Image Version History

Understanding the cadence of OpenAI's image releases helps calibrate expectations for what comes next.

One detail worth noting: `chatgpt-image-latest` is a separate API alias that points to "the image snapshot currently used in ChatGPT." A silent ChatGPT-side upgrade could happen behind that alias without OpenAI publicly naming the updated model.

On LM Arena's April 2026 single-image-edit leaderboard, `chatgpt-image-latest-high-fidelity` ranks first — above `gpt-image-1.5-high-fidelity` in fifth place. That gap is real. It doesn't prove a secret model exists, but it explains why people keep looking for one.

The Rumours: "Tape Models" on LM Arena

The strongest current rumour cluster centres on a set of anonymous image models that reportedly appeared on LM Arena in early April 2026.

What News Outlets Reported

On April 6, 2026, news outlets reported that OpenAI appears to be testing an "Image V2" model on LM Arena under three anonymous codenames: packingtape-alpha, maskingtape-alpha, and gaffertape-alpha.

The reports noted those models were later pulled from Arena. They also claimed some ChatGPT users were encountering outputs from these models, either through A/B testing or ongoing limited access. OpenAI has not confirmed any of this publicly.

What Community Testers Observed

Community posts corroborate that something unusual appeared in Arena around April 4, 2026. One user posted that a new OpenAI image model with "very strong world knowledge and text rendering" was appearing on Arena under the three tape codenames.

Another community member posted examples from the same models, praising their world knowledge and text rendering — with outputs involving an engineer's screen, a selfie with Sam Altman, a NeurIPS poster hall, Stanford campus, and a YouTube screenshot.

Other anecdotes describe one tape model allegedly rendering a watch time correctly and beating Nano Banana Pro on a Minecraft-Manhattan rendering task — while still failing a Rubik's cube reflection test.

These are community observations, not primary confirmation from OpenAI or Arena. They should be treated as directionally interesting, not conclusive.

Why the Rumour Is Plausible

Arena's own documentation confirms that image battle mode serves users two anonymous models, that Arena has helped test pre-release models, and that models tested anonymously can later be marked "preliminary" after a public release. A frontier lab quietly testing an unreleased model on Arena is not far-fetched. What isn't verified is which lab owns the tape models, or what the final public name would be.

The rumoured strengths also align with OpenAI's stated direction. OpenAI officially documents world knowledge and text rendering as GPT Image's core differentiators — and admits that current models still struggle with precise text placement, multilingual text, and some structured compositions.

The tape model examples focus on better UI mockups, more legible button text, accurate timestamps, and grounded real-world scenes. That's technically coherent with where OpenAI says they're still improving.

What Would GPT Image 2 Actually Improve?

As mentioned, based on OpenAI's own documented weaknesses in the current GPT Image family, the most likely areas of improvement for any GPT Image 2 model are:

• Text fidelity — more accurate rendering of spelled-out words, UI labels, timestamps, and multilingual strings
• UI and layout rendering — better handling of app screenshots, dashboards, and structured screen compositions
• Real-world scene grounding — more accurate depiction of real locations, events, and branded environments
• Character and face consistency — improved performance on multiple faces and repeated characters
• Complex structured prompts — better handling of prompts that specify precise spatial arrangements.

These aren't speculative wish-list items. They're areas OpenAI specifically lists as current model limitations, and they match exactly what rumour testers highlighted when evaluating the tape models.

GPT Image 2 Release Date: When to Expect It

There is no confirmed GPT Image 2 release date. OpenAI has not announced one. That said, the timeline from gpt-image-1 (April 2025) to gpt-image-1.5 (December 2025) was roughly eight months. If testing of something like the tape models is genuinely underway in April 2026, a release within the next few months would fit that cadence — but this is inference, not a forecast.

One useful calibration: in December 2025, TestingCatalog reported two codenamed models and speculated they would ship as Image-2 and Image-2-mini. The public release that followed six days later was GPT Image 1.5, not a model called Image-2. Outside observers can be directionally right about "something new is being tested" while still getting the public name and packaging wrong. The same applies now.

The honest summary: if the tape models are real OpenAI models currently in Arena testing, release could come anywhere from weeks to months out. If they're something else, or misidentified, the timeline is unknown. Nothing in any official OpenAI source currently points to an imminent gpt-image-2 launch.

Competitive Context: Where OpenAI Stands Right Now

One reason people expect another OpenAI image push: the competitive pressure is visible in public benchmarks.

On the April 9, 2026 LM Arena text-to-image leaderboard, Google's gemini-3.1-flash-image-preview (Nano Banana 2) holds first place. OpenAI's gpt-image-1.5-high-fidelity sits second.

On the single-image-edit leaderboard, OpenAI's chatgpt-image-latest-high-fidelity leads first place, with gpt-image-1.5-high-fidelity fifth.

OpenAI is competitive, especially in AI image editing, but not dominant across the board. Google's Nano Banana 2 leading text-to-image is a meaningful signal. It's exactly the kind of gap that historically precedes a major release.

What to Believe and What to Ignore

The tape models are the strongest signal here — not because they're confirmed, but because the reports are specific, named, and technically coherent with OpenAI's known roadmap. Something unusual appeared in Arena. Multiple independent sources noted the same standout qualities. That's meaningful. What it doesn't tell you is who built those models, what they'll be called publicly, or when they'll ship.

The parts of the story worth ignoring: claims that GPT Image 2 is "imminent," that a mini variant is confirmed, or that release timing is known. None of that appears in any official OpenAI source. The December 2025 cycle — where leaks pointed to "Image-2 and Image-2-mini" and the actual release was GPT Image 1.5 — is a useful reminder that directional accuracy and naming accuracy are different things.

OpenAI is clearly developing a next image model. Whether it ships as GPT Image 2, GPT Image 1.6, or something else entirely, no one outside OpenAI knows yet.

How to Use GPT Image Models Today

While gpt-image-2 is still unconfirmed, the current GPT Image models are available right now — and they're worth trying.

On getimg.ai, you can access both GPT Image 1 and GPT Image 1.5 directly alongside many other leading image models — including Nano Banana 2, FLUX.2, Seedream 5.0 Lite, and Grok Imagine — under one subscription. The platform's auto-selection handles model routing, so you can focus on output quality rather than model research.

When GPT Image 2 releases, it will be added to the platform. Until then, GPT Image 1.5 remains the best publicly available version of OpenAI's image generation technology, and it's available for generation and editing workflows right now. Try it now in getimg.ai's Content Generator.