When we launched Max a year ago, it didn't include Claude Code, Cowork didn't exist, and agents that run for hours weren't a thing. Max was designed for heavy chat usage, that's it.
GENERATIVE AI
-
System prompt improvements enhance model reliability
By
–
The .20 update has a few system prompt improvements that improve reliability across all models. Working on it!
-
AI generates image guides for real slide creation
By
–
You can do anything you want! In this case, I had the AI use the images as a guide to build real slides.
-
Model Search Behavior in Agentic AI Systems Lacks Relevance
By
–
But then the search attempt it makes prior to generating during the Thinking step has zero relevance, and that doesn't sit right with me. The model is already agentic enough to search for the relevant and correct info to pass it to the image model's context, and it's not doing
-
Using Claude Code for Second Opinions on Development
By
–
For the most part so am I, but I love having a second opinion, which is mostly what I’ve been using Claude Code for.
-
Using GPT Image and Codex to improve UI implementation
By
–
GPT Image 2 + Codex: or how to make Codex not suck at UI.
— Peter Gostev (@petergostev) 21 avril 2026
Step 1: Generate a UI image (native in Codex)
Step 2: Get Codex to implement the UI based on it
Step 3: Get Codex to iterate until it aligns with the image as much as possible
Codex is bad at initial UI, but very good at… pic.twitter.com/3ln7vRIWp2GPT Image 2 + Codex: or how to make Codex not suck at UI. Step 1: Generate a UI image (native in Codex)
Step 2: Get Codex to implement the UI based on it
Step 3: Get Codex to iterate until it aligns with the image as much as possible Codex is bad at initial UI, but very good at -
GPT Image 2 Demonstrates Remarkable Accuracy in Creative Generation
By
–
Yeah, GPT image 2 is *that* good. It’s just so freaking accurate. Image: a 20 person horde raid is fighting Sam Altman in 2004 world of Warcraft style. One shortted.
-
OpenAI Image Model Hallucinations Limit Practical Usability
By
–
I’m seeing really good reviews of OpenAI’s new image model, but I find that even with Thinking, it hallucinates quite a few data points, which, therefore, makes the result unusable no matter how readable and well-formatted it is… Am I missing something?
-
Helper images created by code, then turned into worksheets with imagegen.
By
–
Yes. It appears to create what it calls "helper images" by writing code to produce the maze itself, and then turns it into a worksheet using imagegen (adding the title, the name in cursive, etc.)