Google’s Gemini 3 Pro: Better than mid

Welcome to AI Decoded, Fast Company’s weekly newsletter that breaks down the most important news in the world of AI. I’m Mark Sullivan, a senior writer at Fast Company, covering emerging tech, AI, and tech policy.

This week, I’m focusing on gathering some informed opinions from people trying out Google’s new Gemini 3 Pro AI model. I also look at another “circular” AI investment agreement. 

Sign up to receive this newsletter every week via email here. And if you have comments on this issue and/or ideas for future ones, drop me a line at sullivan@fastcompany.com, and follow me on X (formerly Twitter) @thesullivan. 

What smart people are saying about Google’s Gemini 3

The so-called generative AI boom is only about three years old. It has been characterized by some breakthrough moments, chief among them the release of OpenAI’s ChatGPT in late 2022. A relatively small number of AI labs have been competing to release frontier models that beat all others. The top spots on the rankings of benchmark test scores seem to change names every six months or so. But the release of Google’s new flagship Gemini 3 Pro model (in preview), along with its impressive benchmark test scores, seem like a moment we’ll remember. 

Now, people—many of them developers—who immediately began testing the new model are beginning to weigh in on how well Gemini 3 performs in real-world use. Here is a selection of their impressions. 

Thumbs-up

“Gemini 3 . . . shows significant gains in reasoning, reliability in multi-step agent workflows, and an ability to debug tough development tasks with high-quality fixes. In early evaluations, it improved Warp’s Terminal Bench state of the art score by 20%.” –Zach Lloyd, CEO of Warp 

“I simply asked Gemini 3 Pro to diagnose and fix its own code. It reasoned through the problem in exactly the way I would expect from a thoughtful junior engineer…” –game developer Josh English on Medium

“Best creative and professional writing I’ve seen. I don’t code, so that’s off my radar, but for me the vibes are excellent. Intelligence, nuance, flexibility, and originality are promising in that distinct way that excites and disturbs me. Haven’t had this feeling since 11/30/22.” –Brett Cooper on X

“Gemini beat my own previously unbeaten personal test. The test involves a fairly long list of accurate to year information, ordered properly, many opportunities for hallucination and then used to achieve a goal. I need a new test, so yeah – I think Gemini’s impressive.” –Richard Knoche on X

“It’s a great model, as far as LLMs go, topping most benchmarks, but it’s certainly not AGI. It’s haunted by the same kind of problems that all earlier models have had. Hallucinations and unreliability persist. Visual and physical reasoning are still a mess. In short, scaling isn’t getting us to AGI.” –AI skeptic Gary Marcus on Medium

“I found the answer, and it’s actually terrifying (in a good way) … The uptake of Gemini has been wild … We are talking about a model that is delivering rich visuals, deeper interactivity, and ‘agentic vibe coding’.” –CodeToDeploy on Medium

“[We] created the first open-source evaluation framework to test how leading AI models respond to self-harm and mental health crisis scenarios, and the results were alarming,” said Sean Dadashi, cofounder of the AI journaling app company Rosebud. “Gemini 3 [is] the safest AI model we’ve seen yet.”

Thumbs-down

“Thoroughly benchmaxxed (optimized to do well on benchmark tests), very mid model. Makes so much errors, I have strong doubts they are serving the model they ran the benchmarks on.” –Infrecursion on X

“My experience with the Gemini [command line interface] has been dreadful. It craps out at least half of the time. When it works it is ridiculously fast so I keep trying it. But it has proven very inferior to the Claude code experience in my usage.” –Reddit user dinkleberg

“Was much worse than GPT-5.1 for ‘find me research on [x]’-type queries. It kept trying to do my thinking (synthesis) for me, which is not what I want from it. It gave me individual research results if I explicitly asked but even then it seemed to go way less wide than GPT-5.1.” –Robert Mushkatblat on X 

The free-for-all continues: Anthropic takes $15 billion from Microsoft, Nvidia, with strings 

Someone on Twitter (X) said it best (Peter Wildeford): “After the breakup, both Microsoft and OpenAI are seeing different people . . . also Nvidia is sleeping with everybody.”

In other words, after doubling down on its bet on OpenAI, Microsoft has begun investing in other AI model developers too. Anthropic is the most recent.  

On Tuesday, Anthropic announced it would be taking a new $5 billion in investment money from Microsoft and $10 billion from Nvidia. As part of the deal, Anthropic will purchase about $30 billion of compute capacity from Microsoft’s Azure cloud service, which is powered by Nvidia chips. Anthropic says it’s now the first frontier model to be available within all three major cloud services—Microsoft Azure, Amazon’s AWS, and Google Cloud. Anthropic will work with Nvidia to optimize its models to run well on Nvidia chips. In September, Anthropic raised another $13 billion in funding, after which it was valued at $183 billion.

So, as it has done with OpenAI, Microsoft becomes both an investor in, and a major supplier to, Anthropic. In a less direct way, so does Nvidia. It’s the latest example of the kind of “circular” financial arrangements that have become commonplace in the world of big AI. 

In September, Nvidia announced a $100-billion investment in OpenAI, which the chip supplier will pay in installments that are contingent on OpenAI buying a certain number of chips from Nvidia. So Nvidia gets guaranteed chip sales and a 2% share of OpenAI. “These investments might be circular and raise related party concerns, as Nvidia may own shares in a customer that will likely use such funds to buy more Nvidia gear,” Morningstar equity analyst Brian Colello wrote at the time.  

OpenAI struck a similar deal with Nvidia rival AMD in early October. OpenAI agreed to buy large quantities of AMD’s Instinct AI chips on a set schedule over the next decade. If it keeps to the schedule, it’ll get the option of buying a 10% stake in AMD. 

The group of companies pouring billions into the coming AI revolution isn’t really getting any bigger, and many of the participants seem to be placing bets on each other. The big spenders are Nvidia, Microsoft, Oracle, Meta, Google, Amazon, and some big financiers like Softbank. A larger and more diverse set of players would give confidence that the burgeoning AI industry isn’t just a big hype bubble. 

Survey: Trust in AI is breaking along class divides

Edelman is out with a new survey report, titled Trust in AI report, AI Trust at a Crossroads, which finds that humans’ trust of AI isn’t growing quickly, and that trust levels break along class lines. Edelman surveyed 5,000 people in the U.S., Brazil, China, Germany, and the U.K.

More than half of low-income respondents (54%) feel they’ll be “left out and left behind” in the move toward AI, the survey finds. That’s compared to the 44% of middle-income, and the 31% of high-income people. 

The study found a strong connection between higher trust in AI tools and higher usage of them. Low trust in AI stems from worries over how the systems will use — and whether they will protect — personal data. People, especially in developed countries, also worry about how they might be manipulated by AI. 

At work, only a quarter of non-managers use AI tools weekly, versus 63% of managers. Tech (55%) and finance (43%) employees are most open to AI at work, while adoption is lowest in healthcare (28%), education (25%), food & beverage (23%), and transportation (20%). 

Across all countries, 62% of younger people ages 18 to 34 say they generally trust AI, while 57% of people ages 35–54, and 40% of people 55 and older say they trust it. Interestingly, only 40% of 18-to-34 year olds in the U.S. say they trust AI.   

More AI coverage from Fast Company: 

• Gemini 3 may be the moment Google pulls away in the AI arms race
• Misinformation sites have an open-door policy for AI scrapers
• AI browsers need the open web. So why are they trying to kill it?
• A battle against the AI oligarchy is brewing in this wealthy New York district

Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company Premium.

View the full article