☝🏻 OpenAI’s new image models spotted

New image models have appeared on Chatbot Arena under the codenames: maskingtape-alpha, gaffertape-alpha, packingtape-alpha.

Early results show top-tier world knowledge and near-perfect text rendering - in some blind tests already outperforming Nano Banana Pro.

👀 Some users say it’s already rolling out in ChatGPT.
How to test: Generate a celebrity selfie - if it looks like a real-life photo (no “plastic” feel, highly accurate face), you’re likely using the new model.

📸 NeuralZone

☝🏻 GPT-Image 2 sets a new standard in image quality — now available in @PhotoFixerBot

GPT-Image 2, developed by OpenAI (released April 21, 2026), is the latest image model from OpenAI. Like its predecessors, it offers three quality modes: Low / Medium / High.

Medium is the default (used even in paid ChatGPT plans with ~1K resolution) - and it already outperforms Nano Banana 2 / Pro in user preference benchmarks.

Key Strength: High-Quality Text Rendering
GPT-Image 2 excels at rendering clear, coherent long-form text directly within images - something most models still struggle with.
We tested it using this prompt (from OpenAI website):

an editorial magazine page about wolves in north america and how they're more harmless than we think. make it look like a glossy, smooth, well laid out widely distributed science magazine.

👉 Result [4k hq image]:
• A complete, readable article
• Polished editorial layout
Nano Banana 2 / Pro, using the same prompt, produced only a rough layout with incoherent text.

Detail Test
Prompt:

A highly detailed, high-quality square image of a 5×5 grid of objects, all starting with the letter ‘A’

Result: Image fidelity scales predictably with quality level, with HQ delivering noticeably sharper detail and refinement (see attached images).

Where to try free:

1️⃣Chatgpt.com - low to medium quality, ~1K resolution, with daily limits.

Where to try paid:

1️⃣Paid Chatgpt account - from $8/month.
Provides higher usage limits, but still capped at ~1K–1.5K resolution and Medium quality.

2️⃣Official OpenAI API (for developers)
Up to ~$0.70+ per 4K HQ image.
According to official pricing and the OpenAI's token calculator: a 2880×2880 HQ image consumes ~24K tokens.
At $30 per 1M output tokens, the cost is ≈ $0.71 (+ input tokens for prompts and references).

3️⃣ @PhotoFixerBot
With a Premium subscription, GPT-Image 2 in 4K HQ costs only €0.20 (≈65% cheaper than the official API).

📹 Standard image enhancement is still free in @photofixerbot!

📸 NeuralZone

☝🏻 ChatGPT now has a new default AI model — GPT-5.5 Instant

OpenAI has officially rolled out GPT-5.5 Instant as the new default model for all ChatGPT users, replacing GPT-5.3 Instant.

What’s improved:

🧠 Fewer hallucinations
OpenAI reports a 52.5% reduction in false claims across complex domains like medicine, law, and finance.

📝 Shorter, more precise responses
GPT-5.5 Instant is less verbose, avoids unnecessary formatting, and better understands when a short answer is preferable.

🔎 Smarter web search usage
The model is better at deciding when to use web search for up-to-date information.

🖼 Better image understanding and STEM performance
OpenAI highlights improvements in image analysis, mathematics, science, and technical reasoning.

🧩 More personalized replies
ChatGPT now makes better use of memory, past chats, uploaded files, and connected apps like Gmail (for Plus/Pro users).

😺 GPT-5.5 Instant is already available to all ChatGPT users

📸 NeuralZone