Statistics

In April 2026, Stanford HAI released its ninth annual AI Index Report – the most comprehensive overview of the state of the AI industry. The headlines are predictably optimistic: corporate adoption reached 88%, generative AI outpaced the internet in diffusion speed, and the consumer value of GenAI tools is estimated at $172 billion annually. But look behind the numbers – and a familiar problem emerges.

The AI model market in 2026 works differently. The cost per query ranges from $0.0001 to $0.17 – three orders of magnitude. And the actual quality difference between the top ten models? 0.24 points on a five-point scale. Meanwhile, Wharton / GBK Collective reports that a third of corporate AI projects never get past the pilot stage. And Epoch AI shows that only 5.6% of users apply AI in any genuinely deep way.

Maybe the question isn’t which model is best, but whether paying a premium delivers proportionally better results for typical management tasks.

We tested it. The answer was harsher than we expected.

In November 2025, Stanford data surfaced an unexpected fact: Americans were using AI more at home than at work. We unpacked that in a mysummit.school piece here – and the takeaway then was blunt: work AI was still catching up to personal AI.

Four months later, Epoch AI teamed up with Ipsos for new research – and the picture has shifted.

In March 2025 at SXSW EDU, strategic foresight advisor Sinead Bovell delivered a talk on AI and the future of education. No hype, no panic. But with two studies that change how you should think about AI’s role in learning.

First: a group of students who used ChatGPT without restrictions scored 17% worse than the control group working from a textbook. Second: a different group, where AI was deployed within a fully redesigned instructional system, outperformed a traditional lecture by a factor of two.

Same tool. Opposite outcomes. The difference is in the approach.

This week, Sber unveiled GigaChat Ultra – a new flagship model with a reasoning mode (Thinking). The model is available for free via web, mobile apps, and a Telegram bot. We immediately added both variants to our AI model research for managers: ran them through all 32 scenarios using our unified methodology, scored them with both LLM judges, and compared against the other 52 models.

We conducted a large-scale study: 52 models, evaluations from two independent LLM judges, across 8 categories of management tasks. This is the most comprehensive Russian-language AI ranking for managers available today.

The question remains the same: which AI actually works for a manager in Russia – without VPN, without workarounds?

A Gallup and Walton Family Foundation survey (2024–2025, representative sample of US teachers) produced an impressive number: teachers who regularly use AI save an average of 5.9 hours per week – the equivalent of six full work weeks per school year. Sounds like a solved problem.

But a parallel Royal Society of Chemistry survey (2024, UK) paints a different picture: 44% of teachers tried AI, yet only 3% reported a real reduction in workload. A maths teacher from Ireland explained the gap more precisely than any statistic: “AI generates worksheets quickly, but they need thorough checking – and the time savings turn out smaller than expected.”

Who is right? We previously examined the AI crisis in education from the student side – 86% of students use AI, yet critical thinking is declining. Now – the instructor side. Over the past two years, enough experimental data has accumulated to answer this question with numbers, not opinions.

A year and a half ago, I wrote a note on my personal blog about something I was noticing in my colleagues’ work and in my own: the more you trust AI, the less often you ask yourself “is this actually right?” I was drawing on a Microsoft study at the time – it showed that trust in AI suppresses critical evaluation of the answers it produces. The argument felt strong to me, but it had an obvious flaw: correlation, not causation.

In February 2026, Anthropic researchers Judy Shen and Alex Tamkin published an experiment that closed that gap. Randomized control. Concrete data. And a conclusion that, I think, most people who’ve read about it have misunderstood.

Because this isn’t a story about AI making us dumber. It’s a story about how exactly we use it.