Axios AI+

November 25, 2025
Editor Megan here. Ina is out this week, but I'm here to bring you this short week of AI news in between struggling to figure out if this Macy's Thanksgiving Day Parade French Bulldog Float is real or not.
Today's AI+ is 1,186 words, a 4.5-minute read.
1 big thing: OpenAI faces toughest challenger yet
The world's most popular chatbot, ChatGPT, faces new threats from its biggest competitor: Google's Gemini.
Why it matters: Google was caught on the back foot when OpenAI released ChatGPT three years ago. With the release of, and rave reviews for, Gemini 3 Pro, the script has flipped.
The big picture: Google's new Gemini 3 model is forcing a reckoning at OpenAI.
- CEO Sam Altman told staffers to brace for "rough vibes" and "temporary economic headwinds" as the company works to catch up, per The Information.
- Tech history is full of toppled incumbents — Betamax, AltaVista, MySpace, Friendster — but the AI race moves at a far faster clip. Today's leader could be tomorrow's laggard.
- Even before Gemini 3, OpenAI was already confronting declining engagement, Sources.news reported, as content restrictions designed for user safety squeezed consumption.
State of play: On Nov. 18, Google released Gemini 3 Pro, the latest version of its AI model that will power the company's core search engine and the Gemini app.
- Analysts, users, and industry insiders say Gemini 3's superior benchmarks, integration into Google's ecosystem, and cost efficiencies are pressuring OpenAI, especially after GPT-5's underwhelming August release.
- After spending two hours using Gemini 3, Salesforce CEO Marc Benioff posted on X: "I'm not going back. The leap is insane — reasoning, speed, images, video… everything is sharper and faster. It feels like the world just changed, again."
Reality check: Not everyone is as enamored with Gemini as Benioff.
- "Google is unmatched at the data [it] can train on," Shanea Leven, ex-Googler and current CEO of Empromptu.ai, told Axios in an email.
- This means the model is trained on a wider array of specialized topics. But when Gemini doesn't know about a topic, Leven says she finds it much more willing than ChatGPT-5 to hallucinate an answer.
Zoom out: Generative AI arguably began with Google's 2017 Transformer paper. Much of the technology underlying OpenAI and Anthropic's models traces back to Google, and many current and former researchers at both companies started their careers there.
- Google has nearly every structural advantage: vast revenue and cloud scale, plus the resources to distribute new AI features to billions of users overnight.
- The company is also one of Nvidia's few real competitors in terms of creating its own chips.
- The biggest surprise about Google's rapid gains is how long they took.
Yes, but: OpenAI retains strong brand loyalty from a user base of around 800 million weekly active users.
- OpenAI has been adding memory features to ChatGPT that allow it to give users answers customized to their preferences and prompt history. It's possible — but not easy — to export that history from ChatGPT and import it into Gemini.
What we're watching: Gemini 3 now leads many benchmark tests and could extend that lead when Google's enhanced reasoning mode Gemini 3 Deep Think becomes widely available.
2. OpenAI adds ChatGPT shopping research tool
OpenAI is giving ChatGPT a holiday upgrade with a new shopping research feature that scours product pages, reviews and prices ahead of Black Friday, Cyber Monday and the year-end buying blitz.
Why it matters: Shoppers already turn to ChatGPT to find and compare products, but OpenAI says the new tool delivers deeper, more personalized buying advice than quick specs or price checks.
Driving the news: OpenAI announced yesterday that shopping research is starting to roll out on mobile and web for logged-in ChatGPT users on Free, Go, Plus and Pro plans.
- "To help with holiday shopping, we're making nearly unlimited usage available to all plans through the holidays," the company said in a blog post.
How it works: Ask a shopping question and ChatGPT will suggest shopping research automatically. You can also select "shopping research" from the (+) menu.
- The shopping research feature lets users describe what they need and ChatGPT builds a customized buyer's guide.
- It asks clarifying questions, incorporates past conversations, scans trusted retail sites and pulls in up-to-date details like specs, prices, reviews and availability.
Yes, but: Simple shopping questions — like checking a price or confirming a feature — will still get a regular ChatGPT response.
Between the lines: OpenAI stresses that chats aren't shared with retailers, and results come from "high-quality, publicly available" sites — not ads.
Reality check: The model can still get details wrong, including pricing and availability, OpenAI warns.
- Axios used the new tool to find the best price on a pair of Ugg slippers. We found a price with a URL for $110. ChatGPT was unable to check out through the app and when we went to the website, the lowest price was $150.
What's next: You can click through to retailers to buy today, but OpenAI says direct purchasing inside ChatGPT is coming for merchants that join its Instant Checkout program.
3. Trump boosts AI research to curb energy costs
President Trump signed an executive order yesterday aimed at boosting AI research and development, with an eye toward reducing Americans' spiraling energy costs.
Why it matters: The Trump administration seeks to ensure that government stays out of the way on AI regulation while actively supporting private-sector innovation.
- At the same time, administration officials are eager to address consumer complaints that are mounting over energy bills, job displacement and other economic worries.
Driving the news: The "Genesis Mission" seeks to encourage government information sharing with industry, academia and other scientific institutions.
- Under the order, the Department of Energy will build a platform with AI capabilities for scientists and engineers to use in their work.
The big picture: Though they didn't give any cost estimates, administration officials portrayed the effort as the largest marshaling of federal scientific resources since the Apollo space program in the 1960s.
- "The Genesis Mission will use AI to automate experiment, design, [and] accelerate simulations and generate predictive models for everything from protein folding to fusion plasma dynamics," Michael Kratsios, who heads the White House Office of Science and Technology Policy, told reporters.
Energy Secretary Chris Wright, who touted his agency's national labs' role in the project, argued that AI can help bring down costs.
- "The ultimate goal of this is to make the lives better for American citizens," including creating job opportunities, he said at a press briefing.
- "In the energy space, it's to bring more energy on, make our electricity grid more efficient and reverse price rises that have infuriated American citizens."
What's next: White House officials contend that the Genesis Mission will usher in major scientific advances.
- "This will shorten discovery timelines from years to days or even hours," enabling scientists to test hypotheses and make currently unreachable breakthroughs, Kratsios said.
4. Training data
- Anthropic's latest Opus 4.5 model integrates with Google Chrome and Microsoft Excel. (TechCrunch)
5. + This
Nothing more disturbing than asking Google's Nano Banana Pro to make great art "more cheerful." h/t Ethan Mollick.
Thanks to Matt Piper for copy editing this newsletter.
Sign up for Axios AI+








