Just like AI models, AI news never sleeps.
Every week, we’re inundated with new models, products, industry rumors, legal and ethical crises, and viral trends. If that weren’t enough, the rival AI hype/doom chatter online makes it hard to keep track of what’s really important. But we’ve sifted through it all to round up the most notable AI news of the week from the heavyweights like OpenAI and Google, as well as the AI ecosystem at large.
As of this writing, the popular AI leaderboard LMArena ranks Gemini 2.5 Pro as the model to beat, followed by ChatGPT 4o, and Grok-3 Preview.
This week, OpenAI and Google continue to try and one-up each other with new model announcements, Nvidia is building supercomputers in the U.S., and LLMs can potentially help us communicate with dolphins.
OpenAI news: Meet GPT-4.1, o3, and o4-mini
OpenAI had a big week. On Wednesday, it launched o3 and o4-mini, the latest generations of its chain-of-thought reasoning models, which can tap into all the available tools in ChatGPT. The o3 model’s agentic capabilities have also made it worryingly good at geoguessing locations based on images alone. Mashable tried it, and the privacy implications are frightening.
Earlier in the week, OpenAI launched GPT-4.1 for its developer API, which it says outperforms GPT-4o and has improved coding and instruction following. To that end, OpenAI is phasing out GPT-4.5 from its API (yes, the one that just launched in February). GPT-4.5 will still be available in ChatGPT. Confused about all the different model names and what they do? CEO Sam Altman is aware, and he’s previously said that the company is trying to do a better job of “simplifying our product offerings.”
As OpenAI keeps churning out new models, there are reports that the rapid-fire deployments have come at the expense of safety testing. Testers reportedly only have days to conduct evaluations, according to the Financial Times, and GPT-4.1 shipped without a safety report, as TechCrunch pointed out.
Also, ChatGPT now has an image library, so you can store all of your AI-generated images of action figures, dogs portrayed as humans, and Studio Ghibli copycats in one place.
Perhaps building on growing demand to generate and share ChatGPT creations, OpenAI might be working on a social media network or feed to compete with what X does with viral Grok responses, according to The Verge.
Gemini news: Gemini 2.5 Flash and Google’s dolphin communicator
Credit: Andriy Onufriyenko / Getty Images
There’s a recurring theme in AI news: when OpenAI launches a bunch of stuff, Google swiftly follows. So if it’s a big week for OpenAI, it’s usually a big week for Google, and this week was no different. On Tuesday, Google shared that its video generator Veo 2 is now available to paying Gemini Advanced users and in Whisk, the company’s experimental image editing app.
On Thursday (the day after OpenAI’s o3 and o4-mini launch), Google brought a lightweight version its own reasoning model, Gemini 2.5 Flash, to the standalone Gemini app. Gemini 2.5 Pro, its most powerful model, is only available to Gemini Advanced users. Google also got dinged for lacking details about its safety evaluations with the Gemini 2.5 launch, per TechCrunch.
Google also announced that Gemini Live’s screen sharing and camera vision tool is now free to all Android users with the Gemini app.
And now, with the powers of AI, Google can play Dr. Doolittle. In collaboration with Georgia Tech researchers and the Wild Dolphin Project, Google developed a language model that they say can communicate with dolphins. The model, called DolphinGemma, trained on a database of dolphin vocalizations like whistles, squawks, and clicks in order to help researchers better understand dolphin-speak and eventually talk back to the majestic sea mammals.
This Tweet is currently unavailable. It might be loading or has been removed.
Nvidia, Anthropic, and Grok news
OpenAI and Google often dominate the news cycle, but Nvidia also had big — supercomputer big — news to share this week. On Monday, it announced plans to manufacture AI supercomputers in Texas and build and test its coveted Blackwell chips in Arizona. Over the next four years, the company plans to invest $500 billion in AI infrastructure in the U.S.
The move to develop AI hardware and infrastructure in the U.S. is undoubtedly the result of President Donald Trump’s tariffs, particularly in Taiwan, where Nvidia’s semiconductor manufacturer Taiwan Semiconductor Manufacturing Company operates. Nvidia’s U.S. manufacturing efforts will still involve TSMC, as well as chipmakers Foxconn and Wistron and semiconductor packagers Amkor and SPIL.
After some whiplash tariff back-and-forth, the economic uncertainty and looming trade wars with China are likely Nvidia’s main factor in “hardening supply chain resilience” by building in the U.S., as the press release describes. Either way, it’s a win for President Trump, and for Texas.
In other news, Anthropic announced a Claude integration with Google Workspace, meaning the AI assistant can read your emails. Grok now has a memory and something called Grok Studio, which is a new interface for working on projects within the app.
And last but not least, everyone’s favorite benchmarking platform Chatbot Arena is becoming a real company, Bloomberg reports. In a blog post, the company’s founders wrote, “We are starting a company to support LMArena! LMArena will stay neutral, open, and accessible to everyone. We will focus on improving our open community platform for testing and evaluating LLMs.”
