Ai achievements
Everything here is “generative ai”. Transformers, diffusion models, self attention, or GANs.
Here.
Medicine, Chemistry, and Biology:
LLM specific:
https://www.anthropic.com/news/claude-discovers-novel-enzyme-system
https://openai.com/index/accelerating-biological-research-in-the-wet-lab/
https://www.nature.com/articles/s42256-026-01298-6
https://openai.com/index/diagnose-rare-childhood-diseases/
Paper: https://ai.nejm.org/doi/full/10.1056/AIcs2501343?query=featured_home
https://www.anthropic.com/research/Claude-accelerates-protein-design
https://www.nature.com/articles/s41467-026-76760-5
“We used Chai-1 with ESM protein language model embeddings as input29.”
https://www.nature.com/articles/s41586-025-10097-9
“powered by large language models,”
https://www.nature.com/articles/s41587-024-02143-0
“Here we used PandaOmics, a commercially available target-discovery platform using multiple AI engines including generative pretrained transformers14”
https://www.science.org/doi/10.1126/science.adz4433
“Across a variety of scenarios and applications, the large language model outperformed both human physicians and older models, suggesting its potential utility for clinical care. “
Other forms of ai:
https://www.drugdiscoverytrends.com/ind-approval-ai-designed-usp1-inhibitor/
Ai assisted brain surgery that actually happened on a person: https://www.bbc.co.uk/news/articles/cjwg5n7y68xo
https://www.nobelprize.org/prizes/chemistry/2024/press-release/
https://alphafoldserver.com/welcome
uses the diffusion architecture (same thing as “ai art”)
(paper: https://www.nature.com/articles/s41586-024-07487-w )
can see “Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture “
https://www.biorxiv.org/content/10.64898/2026.10.01.756017v1
deepmind designed enzymes from scratch with a diffusion model (again same thing as ai art) that do chemistry nature doesnt, and they tested them in an actual lab
“de novo enzyme design can outperform natural sequence mining in addressing challenging chemistry”
https://edition.cnn.com/2026/07/16/tech/mayo-clinic-ai-healthcare
https://www.science.org/doi/10.1126/science.aec2657
https://www.nature.com/articles/s41591-025-03743-2
https://www.nature.com/articles/s41591-025-04042-6
https://www.nature.com/articles/s41586-025-09298-z
https://www.nature.com/articles/s41586-025-09721-5
Weather forecasting: https://deepmind.google/science/weathernext/
Physics:
https://journals.aps.org/prxintelligence/abstract/10.1103/9qlr-jp6x
https://journals.aps.org/prxintelligence/abstract/10.1103/dldf-7tfm
(again uses the diffusion architecture)
https://www.nature.com/articles/s41467-024-49104-4
https://link.springer.com/article/10.1140/epjc/s10052-024-13701-w (GANs)
https://journals.aps.org/prc/abstract/10.1103/nmmz-bwcs (also diffusion)
https://www.nature.com/articles/s41467-026-72292-0
https://www.nature.com/articles/s41586-026-10952-3
Mathematics:
This should be said first, because it’s the most important, a literal millennium prize problem: https://openai.com/index/navier-stokes-solution/
Over a hundred problems being solved: https://github.com/openai/math
overview of everything in it: https://github.com/openai/math/blob/main/overview.pdf
summary of the biggest ones (it claims to have solved 90 of the 500 most important open problems in math): https://x.com/aisafetymemes/status/2107622121618964829
some of the most significant ones:
quasi riemann hypothesis, proves the riemann zeta function has no zeros with real part above 7/8. this one is formalized in lean, so anyone can machine check it: https://github.com/openai/math/blob/main/preprints/The-Quasi-Riemann-Hypothesis-September-30-2026/paper.pdf
hilbert’s tenth problem over the rationals, proves no algorithm can decide whether a polynomial equation has a rational solution: https://github.com/openai/math/blob/main/preprints/Hilberts-tenth-problem-over-the-rational-numbers-September-24-2026/main.pdf
Onto others
https://www.erdosproblems.com/1196
https://chatgpt.com/share/69dd1c83-b164-8385-bf2e-8533e9baba9c (chat link used to solve the problem).
Just see the chat in the chat link. Merely a prompt statement of the unsolved problem with no human ingenuity, and the ai model was the one which solved the unsolved problem.
This result has been looked at by several mathematicians including Terence Tao, who is the greatest living mathematician, and they uploaded a paper on it: https://arxiv.org/abs/2605.00301
Where they literally say “We introduce a new method for bounding Erdős sums of primitive sets, suggested from output of GPT-5.4 Pro, based on Markov chains with von Mangoldt weights.”
Keyword: “new”
thought thats a more minor problem. these are greater discoveries:
https://www.anthropic.com/research/riemann-zeta
https://openai.com/index/model-disproves-discrete-geometry-conjecture/
https://x.com/__eknight__/status/2075643450196971805 (cycle double cover conjecture: https://arxiv.org/abs/2607.16356)
https://x.com/EdgarDobriban/status/2077082912021786660
https://en.wikipedia.org/wiki/Jacobian_conjecture#Counterexample_for_n_%3E_2 (jacobian conjecture)
Another math problem solved by Alpoge (same guy who did the jacobian conjecture) with Claude above: https://alpo.ge/s6.pdf
This one is even more significant because it reportedly contradicts already peer reviewed and published math papers and finds the flaw in the previous papers.
If it was just literature searching, it would take that publication as gospel, since that’s what academia treats peer review as. But It didn’t.
https://x.com/DmitryRybin1/status/2079904005652893709 (Dinitz-Garg-Goemans conjecture)
https://cdn.openai.com/pdf/ten-proofs-oai.pdf
https://cdn.openai.com/pdf/51126fac-1b68-4128-9666-c908bcc16033/long_gaps.pdf
And a bunch of more: https://vibemathed.com/
Cybersecurity/coding:
https://arxiv.org/html/2610.06783v1#Sx1.SS0.SSS0.Px1
“An Anthropic employee used an internal research model to investigate open problems in the theory of cryptography.”
This paper is significant because it breaks two strong hypotheses that computer scientists believed was a big barrier. It breaks them both and shows that its actually not a barrier. Even though it looks incremental, it’s insane. Imagine if it was something like if someone found out how to go 1.000004x the speed of light. Very little improvement, but crazy they were able to achieve past speed of light in the first place.
https://red.anthropic.com/2026/cvd/
anthropic’s vulnerability disclosure dashboard, real bugs in real open source projects, not benchmarks
“As of October 2, 2026, we’ve disclosed 6,157 vulnerabilities across 591 open source projects. To our knowledge, 516 of these have been patched.”
You can’t google solutions to novel problems that not even these humans can do mid competition.
Literally beat the greatest competitive programmer in the world, Gennady Korotkevich, by a pretty big margin.
https://red.anthropic.com/2026/mythos-preview/ “During our testing, we found that Mythos Preview is capable of identifying and then exploiting zero-day vulnerabilities in every major operating system and every major web browser when directed by a user to do so. The vulnerabilities it finds are often subtle or difficult to detect. Many of them are ten or twenty years old, with the oldest we have found so far being a now-patched 27-year-old bug in OpenBSD—an operating system known primarily for its security.”
no one knew of this zero day vulnerability in openbsd (one of the strongest and secured OS) before
Glm 5.3 doing a bunch of exploiting as well: https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities
This one is special. It’s not just “publicly available to use”, it’s even open weights. You can straight up download it and run it on your own hardware instead of using a company or being connected to the internet. https://huggingface.co/zai-org/GLM-5.3
And even more exploiting: https://openai.com/index/path-to-astra/
And note, that all of this intelligence is transferable. LLMs are not like specialized humans that only get a degree in one or two fields. They do everything. Being good at math does not mean just doing math. That means very good at logical reasoning in general, which results in transferable gains everywhere, like coding, biology, chemistry, engineering, and more.