Quick answer
On 20 November 2026 Levent Alpöge, a mathematician at Anthropic, released a proof of the Riemann Hypothesis that he says he developed with Claude over about six weeks. The proof is written in Lean, a language computers can check step by step, and runs to about 17 million lines. The next day the Lean and Mathlib communities said they had verified it and have “full confidence” it is correct. It is one of the most famous unsolved problems in maths, first posed in 1859.
Key takeaways
- A 167-year-old problem about how prime numbers are spread out has a computer-checked proof, developed by a human mathematician working with Claude.
- The proof was verified by computer, using several independent proof checkers, not just read by experts.
- The lesson for everyday AI users: AI is most trustworthy when its work can be checked. Ask for answers you can verify.
- GLM 5.5 has been teased by Z.ai, but the idea that it is the mystery Gestalt model is only a guess.
- ChatGPT Space is rolling out to Plus users, and Google showed Stitch Loop, an agent that keeps improving a product towards a goal.
🛠️ Practical guides
Ready to put this into practice? These step-by-step guides answer two common follow-up questions:
- How to Ask AI for Answers You Can Check in Minutes — simple ways to get AI answers you can verify quickly, with sources, workings and tests, borrowed from how the Riemann proof was checked.
- How to Branch an AI Chat to Try a New Idea Without Losing Your Work — how to fork a conversation in ChatGPT, Claude and coding tools so you can explore an idea without messing up your main chat.
In this guide
- Watch: the source video
- What did Claude help prove?
- What is the Riemann Hypothesis, in plain English?
- How was the proof checked?
- Why does this matter if you’re not a mathematician?
- Is GLM 5.5 coming, and is it the Gestalt model?
- What did Google announce?
- What changed at OpenAI?
- What else happened in AI this week?
- What is confirmed and what is only claimed?
- What should you do about this news?
- Frequently asked questions
- The bottom line
- Related posts
Most weeks, AI news is about new models and features. This week it was also about mathematics, and not in a small way. Here is what happened with the Riemann Hypothesis, what it does and doesn’t mean, and the rest of the week’s AI news.
This round-up is based on the 21 November news video from WorldofAI, with the main claims checked against original sources and news reports where we could find them.
Watch: the source video
What this video covers: WorldofAI’s AI news round-up for 21 November: the GLM 5.5 teaser and whether the Gestalt stealth model could be GLM 5.5, Levent Alpöge’s Claude-assisted proof of the Riemann Hypothesis, Google’s maths work, Claude’s mobile app refresh, Google Loop, Thinking Machines’ Inkling 2 Mini, Google’s Gemma explainer, the Antigravity /btw upgrade, Gemini 3.5 Transcribe in the API, AI GIFs in Gemini, Grok Build 2.0 with Cursor, ChatGPT Space for Plus users, OpenAI’s medical research work, ChatGPT desktop animations, OpenAI staff departures, Nano Banana Pro 2 and Neuralink.
Video by WorldofAI (@intheworldofai). The video was sponsored by Lumo, Proton’s private AI assistant. We have no link with the sponsor.
What did Claude help prove?
Alpöge posted a note on his website saying Claude had helped him prove the Riemann Hypothesis, and released the full proof as a Lean library. According to the Guardian, he says the work took about six weeks and that near the end “Claude far outpaced me”. Reports say the same approach also extends to the Generalised Riemann Hypothesis, a broader version of the problem.
The scale is unusual. The Lean code is about 17 million lines, and WorldofAI cited reports that a human-readable version would run to more than 250,000 pages. This was not a short, elegant paper; it was a huge, machine-checked construction built by a person and an AI together.
The timing was notable too. The note came out one day after Anthropic listed on the stock market, which Alpöge himself called “horrible” timing.
What is the Riemann Hypothesis, in plain English?
It is a statement about prime numbers, the numbers like 2, 3, 5, 7 and 11 that can only be divided by 1 and themselves.
- Primes look randomly scattered: sometimes close together, sometimes far apart.
- In 1859 Bernhard Riemann linked them to a mathematical function (the zeta function) and suggested that its important “zeros” all sit on one particular line.
- If he was right, mathematicians can say much more precisely how many primes there are below any number, and how far off their estimates can be.
- Many other results in maths were proved on the assumption that it is true. A proof turns those “if” results into certainties.
WorldofAI’s analogy is a sat nav: it says you will arrive in 25 minutes and you arrive in 32. The Riemann Hypothesis is a bit like a guarantee about how big that error can ever get, for prime numbers. It is also one of the Clay Mathematics Institute’s seven Millennium Prize Problems, which each carry a $1 million prize.
How was the proof checked?
By computer, which is why mathematicians have accepted it so quickly. In a joint statement the Lean and Mathlib communities said they checked four things:
- The statement being proved is the correct, expert-vetted version of the Riemann Hypothesis.
- The code compiles without errors, which dozens of people reproduced.
- The theorem proved is exactly that statement, confirmed with two independent comparison tools.
- The proof itself passes several independent proof checkers, including some that are themselves formally proven not to accept false proofs.
That approach matters. Nobody needed to trust Claude, or even read all 17 million lines. The work was set up so that a computer could check every step. Mathematicians will still spend a long time understanding the new ideas, and some have voiced unease about big AI companies racing to solve famous problems, but the correctness question has been answered unusually fast.
Why does this matter if you’re not a mathematician?
- It shows AI can help discover new knowledge, not just summarise what we already know.
- It shows how to trust AI work: make it checkable. You can apply the same idea at a much smaller scale. Our guide How to Ask AI for Answers You Can Check in Minutes shows how.
- It doesn’t change your passwords or your computer. Despite links to cryptography in theory, there is no practical effect on everyday security.
- It doesn’t mean AI can be trusted blindly. The proof was trusted because it was checked, not because an AI produced it.
Is GLM 5.5 coming, and is it the Gestalt model?
GLM 5.5 looks likely soon, but the Gestalt link is speculation. Z.ai’s community team reportedly teased that a new model is coming before December. WorldofAI noted it has been about 98 days since GLM 5.3, longer than Z.ai’s usual gap, and that Z.ai has tested models under code names before.
That has led some people to guess that Gestalt, the unnamed model topping Contra Labs’ design benchmark, could be GLM 5.5. WorldofAI is clear that this proves nothing. We covered Gestalt in Has Google Built Self-Improving AI? The SIM Paper, the Gestalt Stealth Model and This Week’s AI News and earlier GLM 5.5 leaks in ChatGPT Wallet, a New Gemini Krypton Checkpoint and the GLM-5.5 Leak: This Week’s AI News.
🔎 Unconfirmed / rumour: Rumour: Gestalt being GLM 5.5, and Nano Banana Pro 2 arriving soon, are unconfirmed.
What did Google announce?
- Stitch Loop. Google Labs’ Stitch Loop is a long-running agent that connects to a codebase and live metrics, works towards a goal such as better security or speed, and submits changes for human approval. WorldofAI described it from an earlier leak (internal name reportedly “Jitro”); check Google’s page for current availability.
- Antigravity /btw. The side-question command can now fork your work into separate threads and bring the best one back into the main task. The same idea works in chat apps; see How to Branch an AI Chat to Try a New Idea Without Losing Your Work.
- Gemini 3.5 Transcribe in the API. Developers can build voice tools that let people edit, rephrase and translate text by speaking.
- AI GIFs in the Gemini app, powered by Gemini Omni 1.2 Flash. For a walkthrough, see How to Make an Animated GIF With AI for Emails and Social Posts.
- Gemma explainer. An interactive site that shows how language models work, including a Gemma 4 tokenizer running in your browser.
- Google and maths. WorldofAI noticed updates to a DeepMind maths repository and wondered if Google has its own result coming. That is guesswork.
What changed at OpenAI?
- ChatGPT Space for Plus users. OpenAI’s shared workspace for interactive pages, presentations and documents is rolling out gradually to Plus users, with mobile editing added.
- Medical research. OpenAI shared work with research labs on brain diseases such as Alzheimer’s and Parkinson’s, plus a new benchmark for medicinal chemistry. Finding a promising molecule is a long way from a safe treatment.
- ChatGPT desktop got small animation touches.
- Staff departures. Reports say six OpenAI employees left within 48 hours, some to Anthropic. Staff moves between labs are common; read little into it yet.
What else happened in AI this week?
- Inkling 2 Mini. Thinking Machines released a 26-billion-parameter open-weight model (only about 3.4 billion active at once) under the Apache 2.0 licence. It can run on a Mac with 32GB of memory. If you want to run models locally, read How Much Memory Do You Need to Run an Open-Weight AI Model?.
- Claude mobile refresh. Claude’s phone app is getting a redesign that makes code, projects and artifacts easier to work with on the go.
- Grok Build 2.0. Code found in a Cursor CLI update suggests a combined Grok Build and Cursor app with access to many models. It isn’t official yet.
- Neuralink. A person with paralysis reportedly played GTA 6 using a Neuralink brain implant, a striking example of assistive technology.
What is confirmed and what is only claimed?
| Story | Status |
|---|---|
| Riemann Hypothesis proof by Alpöge with Claude | Released; verified by Lean and Mathlib communities |
| Proof also covers the Generalised Riemann Hypothesis | Reported; checking continues |
| GLM 5.5 coming before December | Teased by Z.ai; no date |
| Gestalt is GLM 5.5 | Speculation |
| Google Stitch Loop | Announced by Google Labs |
| ChatGPT Space for Plus | Confirmed; gradual rollout |
| Grok Build 2.0 with Cursor | Found in code; not announced |
| Nano Banana Pro 2 | Rumour |
What should you do about this news?
- Ask AI for checkable work. Sources, workings, formulas and tests beat confident answers.
- Be cautious about headlines claiming AI has “solved” things. Look for who checked it and how.
- Try branching your chats when you want to explore an idea without losing a good conversation.
- Wait for official news before switching tools over GLM 5.5 or other rumoured models.
Your next step: next time you ask AI a factual or number-based question, add “show your sources and working so I can check this” and spend two minutes checking one claim.
Frequently asked questions
Did AI solve the Riemann Hypothesis on its own?
No. Mathematician Levent Alpöge says he developed the proof with Claude’s help over about six weeks. It was a human and AI collaboration.
Has the proof been checked?
Yes. The Lean and Mathlib communities said on 21 November 2026 that they had verified the 17-million-line Lean proof and have full confidence it is correct.
Does this affect online security or passwords?
No. The Riemann Hypothesis is linked to cryptography in theory, but the proof has no practical effect on everyday security.
Is GLM 5.5 out?
Not yet. Z.ai has teased a new model before December, but hasn’t confirmed a name or date.
Is ChatGPT Space available on the Plus plan?
It is rolling out gradually to eligible Plus users, so it may not appear on every account straight away.
Was the video sponsored?
Yes, by Lumo, Proton’s private AI assistant.
The bottom line
A proof of the Riemann Hypothesis, built with Claude and checked by computer, is a genuine milestone for AI and for maths. The most useful lesson for the rest of us is not that AI is now a genius, but that AI work becomes trustworthy when it can be checked. Build that habit into how you use AI, and treat the week’s rumours as rumours until they are confirmed.
Related posts
- Has Google Built Self-Improving AI? The SIM Paper, the Gestalt Stealth Model and This Week’s AI News
- How to Fact-Check an AI Research Report Before You Rely on It
- ChatGPT Wallet, a New Gemini Krypton Checkpoint and the GLM-5.5 Leak: This Week’s AI News
