OpenAI’s New Image-Gen Tech Is Now Available for Businesses, and Some Are Already Using It

OpenAI just made its latest image generation model—the one that sparked a viral internet trend of Studio Ghibli-inspired memes—available for software developers to integrate into their applications and products. OpenAI first introduced the image generation model, named gpt-image-1, as a feature in ChatGPT in late March. In just its first week of release, according to OpenAI, 130 million users created over 700 million images with gpt-image-1. The new model boasts several improvements over Dall-E, OpenAI’s earlier family of image-generation models. The most obvious improvement is the model’s ability to accurately render text; previous models struggled to generate text, and would usually return images of gibberish, but the new model is much more accurate and can create clear, legible text. While much of the attention that gpt-image-1 received was due to its ability to simulate anime, it can also create realistic textures, edit parts of an existing image, and create images in various aspect ratios. According to OpenAI, several businesses were given early access to gpt-image-1, including Adobe, which will offer the model within its suite of products, and Airtable, which is using the model to offer auto-translation of marketing materials. Other companies experimenting with the model include Canva, which is using the tech to transform sketches into “stunning graphic elements,” and HubSpot, which is “exploring how OpenAI’s new AI image generation capabilities can help customers create marketing and sales collateral.” On X, OpenAI CEO Sam Altman wrote that the API version of the model also enables developers to set their own moderation sensitivity level. If you don’t want your application to create images of people in bathing suits, for instance, you can adjust the moderation levels. The API also allows developers to control how much the model prioritizes speed versus quality when generating a new image, if the background is transparent or opaque, and the dimensions of the image. OpenAI says that in practice, developers will pay $0.02, $0.07, and $0.19 per generated image for low, medium, and high-quality square images. More information about the API’s pricing scheme can be found here. For businesses looking to expand their creative marketing options without breaking the bank, OpenAI’s new model could be a game-changer. BY BEN SHERRY @BENLUCASSHERRY

I Tested 5 AI Assistants—and What I Found Was Surprising

Recently, the Washington Post invited me to join a blue-ribbon panel of communication experts for an AI writing experiment. Tech reporter Geoffrey Fowler pitched the idea as an old-fashioned bake-off with a modern twist. He asked us to test five popular AI tools on how well they could write five kinds of difficult work and personal emails. Why emails? “It’s one of the first truly useful things AI can do in your life,” says Fowler. “And the skills AI demonstrates in drafting emails also apply to other kinds of writing tasks.” In total, the panel of judges evaluated 150 emails. While one AI tool was the clear winner, the experiment highlighted the benefits of AI writing and communication assistants—and one big limitation. Since we were asked to read all the emails blind, we did not know which were written by ChatGPT, Microsoft Copilot, Google Gemini, DeepSeek, or Anthropic’s Claude. Fowler also had us score emails he had written to see if we could distinguish between AI and a human writer. The best AI writing assistant The clear winner was Claude. “On average, Claude’s emails felt more human than the others,” Fowler noted. Another judge, Erica Dhawan, said, “Claude uses precise, respectful language without being overly corporate or impersonal.” DeepSeek came in second place, followed by Gemini, ChatGPT and, in last place, Copilot. Although Copilot is widely available in Windows, Word, and Outlook, the judges agreed that its emails sounded too much like AI. “Copilot began messages with some variation of the super-generic ‘hope you’re well’ on three of our five tests,” said Fowler. While Claude won this competition, I later learned that my scores showed a preference for the human-written emails. And that’s because all the AI assistants had one big limitation. According to Fowler, “Our five judges didn’t always agree on which emails were the best. But they homed in on a core issue you should be aware of while using AI: authenticity. Even if an AI was technically ‘polite’ in its writing, it could still come across as insincere to humans.” My takeaway: AI tools are great for outline, flow, and clarity of argument. But they’re often stilted, formal, robotic, and lack personalization, emotion, and empathy. AI assistants have trouble with creativity because the architecture on which they’re based (large language models) generates content with “syntactic coherence,” an academic term for stringing sentences together that flow naturally and follow grammar rules. But as you know, rules are meant to be broken. Steve Jobs broke the rules For example, in 1997 Apple’s Steve Jobs launched one of the most iconic campaigns in marketing history. The company was close to bankruptcy, and needed something to attract attention and stand out. Apple’s now-famous television ad—nicknamed “the crazy ones”—featured black and white portraits of rebels and visionaries such as Bob Dylan, John Lennon, Martin Luther King Jr, and others. The marketing campaign is credited with redefining Apple’s brand identity helping to save it from financial ruin. If the writing had been turned over to AI, it wouldn’t have happened. How do I know? Claude told me. “If asked to create a slogan like Apple’s famous campaign in my default mode, I would almost certainly have written ‘Think Differently’ rather than ‘Think Different,'” Claude acknowledges. “My training emphasizes grammatical correctness. The proper adverbial form to modify the verb ‘think’ would be ‘differently,’ and I’d be inclined to follow this established rule.” Claude says it can analyze why the campaign worked “after the fact … but generating that kind of deliberate grammatical rebellion doesn’t come naturally to me.” AI doesn’t have a rebellious streak because—breaking news—it’s not human. Some bots might perform better than others at simulating human qualities in their writing samples, but they don’t have the one thing you have: a unique voice built on years of personal experiences and creative insights. AI is a helper, an assistant. Use it to brainstorm ideas, clarify thoughts, summarize documents, and gather and organize information. Those are all important and time-consuming tasks. But while AI can enhance communication, it shouldn’t replace the communicator. As more people rely on AI assistants to write emails, resumes, memos, and presentations, there’s a real danger that many people will sound alike—corporate recruiters are already spotting this trend. But you’re not like everyone else. You have a unique and powerful story to share. Don’t let artificial voices silence your authentic one. EXPERT OPINION BY CARMINE GALLO, HARVARD INSTRUCTOR, KEYNOTE SPEAKER, AUTHOR, ‘THE BEZOS BLUEPRINT’ @CARMINEGALLO

OpenAI’s New Models Could Be Its Smartest and Most Powerful, Thanks to This Feature

On Wednesday, OpenAI introduced two new artificial intelligence models to the world. The models, named o3 and o4-mini, are both part of the Sam Altman-led company’s “O” series of reasoning models, which means they’re capable of taking time to “think” through how to best answer a query. OpenAI says o3 and o4-mini are “our smartest and most capable models to date,” thanks to one special feature: tool use. In the AI industry, “tools” usually refer to special abilities that can be bestowed upon an AI model, like the ability to write and run code, search the internet, use a web browser, and parse through internal databases. These abilities are what transform AI models into AI agents, and o3 and o4-mini are OpenAI’s first reasoning models with access to these tools. When you ask o3 or o4-mini a question, it will spend a short amount of time thinking through which tools would be most useful for completing the tasks. It then starts a multi-step process to answer the question. For example, when asked to predict how Donald Trump’s proposed tariffs will affect the burgeoning American AI industry, o3 thought for 25 seconds and then delivered a report sourced from recent articles by Time, Reuters, Axios, and Forbes. The report found, in part, that tariffs would make the hardware that powers AI “noticeably pricier … Expect higher upfront costs, a squeeze on smaller AI outfits, and a brand‑new round of supply‑chain gymnastics.” However, there’s a catch here: Because the model’s training data only goes up to June 2024, the model assumed the tariffs in question referred to Trump’s earlier suggestion of 10 percent tariffs across the board and a 60 percent tariff on China. Of course, American tariffs on Chinese imports are now up to 245 percent. This process isn’t entirely new, and should be familiar to anyone who has used ChatGPT’s Deep Research feature, which similarly turns the platform into an AI agent that can scour the internet and create a lengthy report on the topic of your choosing. Deep Research was actually built on an earlier version of the o3 model, but the current version of o3 is much faster, although may return less thorough answers. While o3 is designed to be a research whiz, o4-mini was created to serve as a coding companion. The model has set new benchmarks across several software engineering tests, and “performs especially strongly at visual tasks like analyzing images, charts, and graphics.” In examples, OpenAI researchers showed that both models will double-check answers to math questions, and will explain how they came to their conclusions. On X, Sam Altman opined on the new models’ capabilities, writing that “the ability of the new models to effectively use tools together has somehow really surprised me. Intellectually i knew this was going to happen but it hits different to see it.” Entrepreneurs are only just getting their hands on the new models now, but they could be helpful for automating internal workflows that were previously too complicated, for example. BY BEN SHERRY @BENLUCASSHERRY

Humans Don’t Want Factory Jobs. This $70,000 Robot Could Be the Future of Manufacturing

As the AI revolution edges closer to putting robots next to human at some jobs, it may be time to brush up on sci-fi writer Isaac Asimov’s three “laws of robotics”.* While companies like Tesla, OpenAI, Figure and even Apple are either dabbling in the android field or charging right ahead with developing fully functioning, AI-powered machines like Elon Musk’s Optimus, one company just stole the spotlight by putting its robot on sale. Meet Reachy 2, a $70,000 humanoid robot from Hugging Face, an AI startup. Given that few American humans want to work in factory jobs, the timing may be great. The breakthrough, announced by Hugging Face co-founder Clément Delangue in a post on X, is the first commercially available open-source humanoid robot, news site Decrypt.co notes. In his X post Delangue admits the price of the machine is expensive, but he also pointed out it’s “already in use at Cornell, Carnegie Mellon & major AI labs for robotics research and education.” The open-source feature and the device’s hackability are standout features here: it means people who buy one can easily tinker with the way the robot works, from code through to hardware, and optimize it exactly to match their needs—and it’s all approved by Hugging Face. The New York-based AI-centric company, which is best known as a platform and community for sharing AI-building technology, says that Reachy 2 is a “a versatile, expressive, and open robotic platform designed to explore the future of human-robot interaction, assistive robotics, and AI-driven behavior.” The robot looks vaguely human from the waist up (its lower torso is a tripod-like arrangement that allows the device to pivot) but its system includes stereo vision, microphones and a lidar system for detecting and locating distant objects. The company says it’s useful if you want to “build an expressive assistant, a teleoperated avatar, or a robot that learns from demonstration,” and highlights that it’s really not a consumer-facing product yet. Reachy 2 is more aimed at researchers trying to adapt robot technologies for future uses, like working in factories. But since the device is now on the market, there may be nothing stopping a manufacturing company buying one and testing it on the shop floor. Hugging Face was previously known for its AI services rather than hardware, but it acquired the makers of Reachy 2 on Monday this week—buying a company called Pollen Robotics for an undisclosed amount. Decrypt noted the company explained it was going to pursue physical versions of its AI systems in a separate X post, noting that Hugging Face now believes “robotics could be the next interface for AI” in a way that is “open, affordable, and hackable.” This may set it apart from companies like Tesla, which has bold ambitions for its Optimus machine, which may not be a fully open-source, hackable machine (given the way Tesla currently manages intellectual property rights for the code for its EVs). Hugging Face’s AI robotics ambitions may have kicked off at just the right moment. Though there is broad support for efforts to increase the number of manufacturing jobs in the U.S., a recent survey highlighted that the average American would prefer not to work in that kind of role. Entrepreneur Peter Diamandis’ prediction that “millions, then billions” of humanoid robots are coming sets the scene for companies like Hugging Face to sell boutique humanoid robots to companies large and small who want to sample a bit of the future of robot-centric factory work right now. *Asimov’s three laws… file them away for future arguments around the water cooler: A robot may not injure a human being or, through inaction, allow a human being to come to harm A robot must obey the orders of human beings except where such orders would conflict with the First Law A robot must protect its own existence as long as such protection does not conflict with the First or Second Law (we may be many years away from having to worry about this one, though, as Reachy 2’s cute demeanor demonstrates). BY KIT EATON @KITEATON