BALTIMORE, MD, June 04, 2026 /24-7PressRelease/ — The last three months will likely be remembered as the busiest stretch in AI launches yet. We’ve been given six new flagship text models, at least two new frontier image models, and at least two ground-breaking video models, all released inside a 90-day window.
At Overchat AI, we ran the numbers to see which models received the most real-world adoption. The numbers below are based on usage data from May 18 to May 18, 2026 at Overchat AI — a platform where users have one-click access to every major model. The data covers 1,312,500 users and roughly 8.6 million AI interactions.
Understanding the New Models
Before we jump into the numbers, here’s a quick primer on what each model is.
Model-Developer-Release date
OpenAI GPT-5.5-OpenAI-April 23, 2026
Anthropic Claude Opus 4.7-Anthropic-April 16, 2026
Anthropic Claude Opus 4.6-Anthropic-February 5, 2026
xAI Grok 4.3-xAI-April 30, 2026
DeepSeek DeepSeek V4-DeepSeek-April 24, 2026
Most Used AI Models
Here’s the data from May 18 to May 18, 2026, across roughly 8.6 million chat interactions:
Model-Total Chats-% of Usage
GPT-5.5-432,000-81.2%
Claude Opus 4.7-28,700-5.4%
Claude Opus 4.6-19,150-3.6%
Grok 4.3-15,720-2.9%
DeepSeek V4-10,250-1.9%
Kimi K2.6-5,800-1.1%
In the previous version of this report, published three months ago, we found that 66% of Overchat AI users opted for GPT-5.1. The past three months have been incredibly busy, even by the standards of the rapid release of AI models, with new generations or updates to existing models being released by Anthropic, Google, xAI, Moonshot, Alibaba and DeepSeek. Nevertheless, the share of GPT usage on Overchat AI — which gives equal access to all via a simple menu switch — has increased to over 80%, while all the other models combined account for less than 19%.
This is interesting because you would expect the market to become more fragmented, not less. However, it is clear that GPT is doing something right. This could be because it is the best generalist, whereas other models are more specialised and geared towards specific use cases. Many of these use cases require an agentic style of work.
There are two other things in the data worth pointing out:
– Many people are continuing to use Opus 4.6, even though Opus 4.7 has been released. The older model is 3.5 months behind the new release, yet still captures 67% of usage. This is unusual, as in most product categories, version N+1 tends to replace N within a release cycle. Opus 4.7 has been poorly received, with many people complaining of a decline in writing quality and of it often refusing to complete tasks that Opus 4.6 would do. This could explain the continued popularity of the older model.
– Not a lot of people are choosing Grok 4.3 and DeepSeek V4, despite these models being genuinely impressive.
GPT-Image-2 Has Overtaken Nano Banana Pro
Nano Banana Pro used to be the undisputed leader in image generation. If you wanted to generate a good AI image, you would open Nano Banana Pro without considering any other options. This is no longer the case.
On 21 April 2026, OpenAI released GPT-Image-2. This advanced model can think before generating images and pull live web references to accurately render things it doesn’t recognise. It can also generate up to 4K images. However, all of this comes at a cost: GPT Image is expensive and slow, which you might think would make it situational. But that’s clearly not the case (more on this below).
Google has also released a new model called Nano Banana 2. Despite dropping the ‘Pro’ from the name, it generates images with high visual fidelity at a lower price and faster than the Pro version. So, which do users prefer: GPT-Image-2 or Nano Banana-2? Here’s a table answering that question:
Model-Total Chats-% of Image Usage
GPT-Image-2-1,510,000-43.7%
Nano Banana Pro-1,105,000-32.0%
Nano Banana 2-804,000-23.3%
Seedream 5-36,000-1.0%
It is clear from the numbers above that people are willing to wait longer and pay more for higher quality when it comes to image generation, as people use the slower and more expensive GPT-Image-2 roughly, 1.4× more often than Nano Banana Pro.
The split between Nano Banana Pro (32.0%) and Nano Banana 2 (23.3%) reinforces this. Google’s faster, cheaper sibling did pick up real volume, but it didn’t cannibalize Pro the way you’d expect. Users self-segment by quality budget rather than collapsing to the cheapest option.
Sora 2 Is Still The Most Loved Video Model
In December, we reported that over 80% of users had chosen Sora 2 for video creation. During this measurement period, this figure has fallen to 64%, as it is being replaced by newer models. Specifically, Seedance 2.0 (released in February 2026) is used 29% of the time, while Kling 3.0 (released on 5 February 2026) is used around 7% of the time.
Model-% of Video Usage
Sora 2-64.4%
Seedance 2.0-28.8%
Kling 3.0-6.8%
The Sora consumer app was shut down on 26 April 2026, and the API is scheduled to end on 24 September 2026. Additionally, both Seedance 2.0 and Kling 3 offer a more advanced visual experience.
Bottom Line
To conclude, here are the key findings from 1,312,500 users and ~8.6 million AI interactions, May 18–18, 2026:
– Despite the saturated release landscape, ChatGPT is the most popular AI model by a landslide. During our last measurement period, it accounted for 66% of usage; in this period, however, it accounted for 81%.
– GPT-Image-2 has become more popular than Nano Banana 2 and Nano Banana Pro. Until recently, Google was the undisputed leader in image generation, but OpenAI has changed that by developing a more advanced image model, despite it being more expensive and slower.
– Sora 2 remains the most popular video model, but people are gradually switching to other options. Its usage has fallen from over 80% to 64%, with Seedance 2.0 at 29% as a close second.
About Overchat AI
Overchat AI is an AI platform that brings together over 150 AI tools for image, video, and text generation, helping users solve tasks though AI-powered mini-apps.
We’re trusted by over 350,000 active users, who come to Overchat AI to create and edit images, make AI videos, solve homework tasks, chat with PDF files, and write texts.
Every task on Overchat AI, from combining two images into one to humanizing text, is supported by a specialized interface (a mini-app) that makes it incredibly easy to get the job done. Overchat AI offers a library of 150+ AI tools, which is always expanding.
On top of that, Overchat AI gives users access to the latest AI models for text, image, and video generation, including GPT 5.5, Claude Opus 4.6, Gemini 3.5, Kimi K2.6, and many others. All of these models are included in a single subscription, helping users save tens — or even hundreds — of dollars per month compared to paying for multiple AI services separately. (more on pricing below).
Overchat AI operates on a freemium model, allowing users to try image, audio, and text generation for free.
Free users receive 100 tokens every day, which can be spent on supported tools and AI models.
These tokens refresh automatically every 24 hours.
In addition, Overchat AI offers a wide range of free AI tools and chatbots on the overchat.ai website, giving users plenty of opportunities to explore the platform before upgrading.
Pricing:
Overchat AI Pro is the platform’s premium tier that unlocks full, unrestricted access to all Pro tools and models. Pricing is as follows:
$14.99/month
49.99/year
It operates on a subscription basis, which the user can cancel anytime.
# # #
