OpenAI confirmed the figure of 700 million active users per week. Let me remind you that the plan is to reach the bar of a billion users per day by the end of the year. In January, this seemed like an exorbitant figure, but the trajectory is solid.
Only works for Perplexity Pro/Max subscribers. Uses workaround via Requestly extension.
function modifyRequestBody(args){const { method, url, body, bodyAsJson }= args;
if (method === "POST" && bodyAsJson){const modifiedBody = { ...bodyAsJson };
if (!modifiedBody.params) modifiedBody.params = {};
modifiedBody.params.model_preference = "gpt5";
return JSON.stringify(modifiedBody);
}return body;
}Save Rule (Ctrl+S)
May stop working after Perplexity updates
Today, 17:00 London / 19:00 Moscow (live on Twitter/X). What we know so far: the announcement video resembles Operator, a browser agent that was tested earlier.
What happened:
PS Despite the legal risks, transfers between direct competitors (Anthropic → Cursor, Meta → OpenAI) are happening smoothly.
On July 14, OpenAI will permanently remove GPT-4.5 from the API.
What will replace it? Still unknown. Possible options:
Context:
What should developers do?
Personal opinion. It's unlikely that GPT-5 will be released so suddenly. Most likely, there will first be a "soft" transition to something like an improved GPT-4-turbo, and only then, in the fall, will they start testing version five.
Why This Matters
We are entering an era of model specialization:
Key Features of o3 Pro
A qualitative leap beyond Claude Opus or Gemini 2.5 Pro
P.S. I already have access testing starts now!
What is it?
Sparkify is Google’s upcoming generative video tool where:
Early Access. Join waitlist Sparkify by Google
While OpenAI defends its upcoming corporate restructuring from a range of critics from the nonprofit world, it is likely pleased with its decision last year to hire a number of employees with deep ties to Democrats.
These hires, including Democrat Chris Lehane, who leads OpenAI’s global affairs team, looked like miscalculations when Donald Trump won the presidency, especially given CEO Sam Altman’s previous stance (he reversed himself after the election, donating $1 million to the president’s inaugural fund and working with the administration on the Stargate data center project).
But OpenAI doubled down on Democrats. After hiring Lehane, the Global Affairs team brought in Debbie Mesloh, who advised Kamala Harris during her campaigns for attorney general and senator, and Marisa Moret, formerly managing counsel and chief of staff for the San Francisco city attorney.
OpenAI also hired Ann O’Leary, who previously served as chief of staff to California Governor Gavin Newsom and advised Hillary Clinton during her 2016 presidential campaign, as outside counsel. Ann O’Leary’s law firm Jenner & Block is one of the firms that the President recently issued some nasty executive orders against for hiring lawyers who worked on investigations into him.
While Democrats are out of power in Washington, OpenAI is focusing more on the local level. All of the Democrats listed above are helping OpenAI match the strength of groups opposing its planned restructuring, including a coalition of more than 60 nonprofits, many based in California, that have petitioned the state attorney general to intervene. Critics of the restructuring argue that the nonprofit that owns and controls OpenAI could be short-changed in the restructuring of its for-profit subsidiary.
Fred Blackwell, CEO of the San Francisco Foundation, and Orson Aguilar, president and CEO of the advocacy group LatinoProsperity, are leading the coalition.
Blackwell and Aguilar met with OpenAI representatives in March, accompanied by Julián Castro, who served as Obama’s secretary of housing and urban development. On the other side of the table were Mesloh, Moret, and Daniel Zingale, who have worked for three California governors.
Zingale is assembling an advisory committee that will make recommendations to OpenAI’s board in July on its governance structure and how to pursue its philanthropic mission. The five-member committee includes two former executives of the California Endowment, one of the organizations in the coalition of nonprofits opposed to the takeover, and Dolores Huerta, the legendary labor activist who co-founded the United Farm Workers with Cesar Chavez.
In a letter responding to the coalition’s demands, OpenAI said its advisory committee has reached out to nonprofit critics to get their views “on how nonprofits can achieve real impact in areas such as health care, education, public services, and scientific discovery.” The nonprofit coalition and OpenAI’s advisory committee are scheduled to meet next week.
A new benchmark, VideoGameBench, evaluates how well AI models can play video games with surprising (and slightly disappointing) results.
Key Findings:
Why This Matters:
The United Arab Emirates will provide free ChatGPT Plus access to all ~1.5 million citizens a world-first initiative in AI democratization.
Why It Matters:
This move aligns with UAE’s National AI Strategy 2031, aiming to make AI a core part of governance and daily life.
Full details: Claude Opus 4
Full details: Claude Sonnet 4
Like ChatGPT, Codex performs best when given precise instructions—these strategies will help you get the most out of it:
Create an AGENTS.md file documenting:
Testing & Validation. Define binary success criteria ("works/doesn't work")
Break changes into small chunks:
Current limitations:
Advanced Tactics. When stuck:
Aerial view of an active construction site for one of Microsoft & OpenAI's rumored "Stargate" AI data centers
Key details:
Fun fact: The "Stargate" codename hints at ambitions beyond current AI capabilities - perhaps anticipating future breakthroughs requiring massive compute.
Finally someone did it! A new site now tracks all 200+ falsifiable predictions from the AI 2027 scenario, organized by:
Current stats:
Key limitations:
What's already come true:
Most anticipated 2025 event: Track the progress: AI 2027 Prediction Tracker
ChatGPT has officially become:
Perspective:
According to SimilarWeb data, OpenAI has actually increased its market share among generative AI services over the past year - growing from ~75% to 80% despite fierce competition and new market entrants.
Key Insights. Remarkable resilience - maintained dominance while competitors multiplied. Comprehensive analysis available - report includes:
The full report reveals how AI adoption is reshaping digital ecosystems beyond just the AI service market itself.
Google has rolled out an update to Gemini 2.5 Pro, with the most significant gains seen in web programming benchmarks (check second image). Respect to Google for daring to compare it directly with Claude 3 Opus – while it loses in many benchmarks (first image),it's interesting that both models perform similarly on SimpleQA, a benchmark testing obscure factual knowledge. This might hint at comparable model sizes (though that's a very rough guess).
Currently, two Gemini versions and Claude 3 Opus are tied for first place in the Arena. From my testing:
Concerning Benchmark Changes (vs previous Gemini 2.5):
The model seems over-optimized for web dev/coding/Arena performance at the cost of other capabilities. Worse – Google replaced the old model instead of offering both versions.
Coming Soon?
A "Computer Use" preview (AI agent controlling browsers/computers) appeared on their site – but it's unclear whether it'll launch before Google I/O (May 20-21).
Google has rolled out its new "AI Mode" search experience across the entire U.S. - it's a new tab, similar to Images or Maps, but inside you’ll find something more like ChatGPT or Perplexity: with memory, product cards, and more. Many of you have probably seen it already.
This is a huge leap toward the mass adoption of large language models (LLMs) - and with that, congratulations to all of us.
OpenAI is doing well - according to Similarweb data for April, the ChatGPT website surpassed X (formerly Twitter) in total visits, and by a wide margin: 4.786 billion vs. 4.028 billion, including both web and mobile traffic. If the numbers are accurate, this places ChatGPT fifth globally, behind only Google, YouTube, Facebook, and Instagram.
Interestingly, the daily traffic graph shows a clear pattern tied to the day of the week - a trend commonly seen in productivity and work-related tools. This suggests that a significant portion of ChatGPT usage is linked to work and education, meaning it has a real economic impact, beyond just generating poems and images.
Alibaba has launched Qwen3, its next-gen AI model family, featuring both Dense and Mixture-of-Experts (MoE) architectures. The flagship Qwen3-235B-A22B (22B active params) outperforms even 32B models in speed and accuracy (o1/R1 benchmarks). The lineup also includes 32B, 14B, 8B, 4B, 1.7B, and 0.6B versions—with smaller models distilled from larger ones.
All models support two inference modes (with and without reasoning, similar to Claude 3.7) and were trained on 100+ languages (including Russian & Ukrainian) across 36 trillion tokens.
Try it for free: https://chat.qwen.ai/
Full details: Model Cards
In the late 1990s, investors bet everything on internet stocks, sending the Nasdaq-100 up 500% in three years—only to crash 80% when profits failed to materialize. Today, history seems to repeat itself: AI stocks like Nvidia have soared 800%, fueled by the belief that AI will reshape the world.
But there’s a key difference. The dot-com bubble was pure speculation—Cisco’s peak valuation hit 150x earnings. Nvidia’s rise, while steep, is backed by real profits, with a P/E ratio around 50 (now 35 after recent dips).
So, is this an AI bubble? Not like the 1990s—yet. Valuations are high, but fundamentals, not just hype, are driving growth. The real test? Whether AI companies can deliver long-term profits, not just promises.
Compare AI. Test. Benchmarks. Mobile Apps Chatbots, Sketch
Copyright © 2026 All Right Reserved.