Skip to content
Bud's Wiki

GPT-Image-2 leak: OpenAI image model on LMArena under maskingtape / gaffertape / packingtape codenames

TL;DR

Section titled “TL;DR”

A rumor-tier tweet from @levelsio claiming OpenAI’s next image model — referred to as GPT-Image-2 — has leaked onto a public side-by-side eval surface under three codenames: maskingtape-alpha, gaffertape-alpha, and packingtape-alpha. The poster reports it shows strong world knowledge and text rendering, “possibly better than Nano Banana Pro” (Nano Banana 2 / Gemini 3.1 Flash Image (Google product announcement) family). No paper, no card, no benchmarks; this is a pointer to a leaked-codename sighting so the wiki has something to link from when an official release lands.

Key claims

Section titled “Key claims”
  • Three codename slots — maskingtape-alpha, gaffertape-alpha, packingtape-alpha — are claimed to belong to an unreleased OpenAI image model dubbed GPT-Image-2 [tweet body].
  • The poster’s qualitative read: strong world knowledge and text rendering, possibly ahead of Nano Banana Pro [tweet body].
  • No supporting prompts, comparison images, or third-party verification in the tweet itself; the only attachments are four sample images attributed to the codenamed models [tweet body].

Method

Section titled “Method”

Not applicable — there is no method here. The tweet is a sighting of model codenames on what is implied to be a public LMArena-style head-to-head eval surface. The triple naming pattern (alpha variants under tape-adhesive theme) is consistent with OpenAI’s prior practice of staging multiple checkpoints under shared themed codenames on Arena for blind ranking before reveal.

Results

Section titled “Results”

No quantitative results. The implicit “result” is one qualitative comparison vs. Google’s Nano Banana Pro from the poster, with no reproducible prompts, no win rates, and no Arena scores attached to the tweet.

Why it’s interesting

Section titled “Why it’s interesting”

This is the kind of pointer the wiki should hold but should not over-weight: a single-source rumor naming an unreleased OpenAI image model that — if real — would land directly against the closed image-gen baseline tracked by Nano Banana 2 / Gemini 3.1 Flash Image (Google product announcement) (Nano Banana 2 / Gemini 3.1 Flash Image) and against the open frontier that Qwen-Image-2.0, FLUX.2 [klein]: Towards Interactive Visual Intelligence, and HunyuanImage 3.0 Technical Report all explicitly position against. Worth filing so that when an official GPT-Image-2 card drops, the three codenames have a stable reference. Until then: treat as unverified.

See also

Section titled “See also”