OpenAI’s GPT-5.4 worked with Molecule.one’s Maria AI and its microliter HTE lab to move from an open-ended medicinal-chemistry goal through literature review and proposal generation all the way to physical experiments that improved a Chan-Lam coupling. The workflow surfaced TEMPO (and a cheaper analog) as an additive that raised mean yields on primary sulfonamide substrates; the automated lab executed more than ten thousand reactions across two cycles before human chemists reproduced key conditions at conventional scale.
This is really cool. Bridging the gap between bits and atoms in science.
GPT-5.4 helped drive a medicinal chemistry project from literature review to a validated experimental result.
Paired with http://Molecule.one’s Maria AI and specialized lab, the model proposed an unexpected way to improve a widely used reaction in drug discovery.
Human chemists ran fourteen representative substrate pairs on the bench and saw higher yields in eleven cases, with most showing more than a twofold jump. The improvement appeared consistent across the majority of boronic acids and sulfonamides screened.
The demonstration stayed within one reaction class, required a specialized high-throughput lab, and kept humans in the loop for steering and final validation. Broader generalization to other transformations or manufacturing settings has not been shown.
Many users praise GPT-5.4 for accelerating validated drug discovery in chemistry with MoleculeOne, while others accuse OpenAI of ignoring GPT-4o's strengths and pushing an inferior model.
No Digg Deeper questions have been answered for this story yet.
GPT-5.4 for improving a challenging reaction in medicinal chemistry:
GPT-5.4 helped drive a medicinal chemistry project from literature review to a validated experimental result.
Paired with http://Molecule.one’s Maria AI and specialized lab, the model proposed an unexpected way to improve a widely used reaction in drug discovery.
GPT-5.4 helped drive a medicinal chemistry project from literature review to a validated experimental result.
Paired with http://Molecule.one’s Maria AI and specialized lab, the model proposed an unexpected way to improve a widely used reaction in drug discovery.
The reaction is Chan-Lam coupling, a method chemists use to build pharmaceutically relevant molecules.
One difficult version, involving primary sulfonamides, has historically produced low yields, limiting how useful it can be in medicinal chemistry.
https://openai.com/index/ai-chemist-improves-reaction/
GPT-5.4 reviewed scientific literature, generated and ranked research proposals, helped design experiments, analyzed results, and proposed follow-up studies.
Human chemists steered the work, selected proposals for testing, and validated the final result.
The full process took about 2.5 months, plus another half month for human chemists to write up the results.
This is an early example of frontier models supporting more of the scientific research loop: reviewing studies, proposing hypotheses, designing experiments, interpreting data, and surfacing findings that human experts can validate.
Maria tested the idea across 10,080 reactions, and human chemists later validated representative results by hand.
Under the optimized conditions, yields improved for 88% of the boronic acids and 83% of the sulfonamides tested.
Human chemists then repeated 14 representative reactions by hand: 11 showed higher yields, including 8 with a more than twofold improvement.
@OpenAI It's the modern version of the serendipity that gave us penicillin, but instead of a moldy petri dish, it's GPT-5.4 doing the heavy lifting. We're moving from accidental discovery to engineered luck.
Here we go again... 📌 GPT-4o 10 June 2026 research : 1,000 participants. Two randomized experiments. Significant improvement in patient understanding.
Published in JMIR Human Factors. Peer-reviewed. Clinical trial registered.
Source : https://humanfactors.jmir.org/2026/1/e89451
🚨GPT-4o clocked a staggering 90.0% top 5 accuracy, completely outperforming specialized, heavily funded deep learning architectures built for that sole purpose. Source : https://pmc.ncbi.nlm.nih.gov/articles/PMC12362839/ 🚨GPT-4o utterly demolished it with an 88.8% accuracy score, showing elite proficiency across surgery, psychiatry, and infectious diseases. Source : https://pmc.ncbi.nlm.nih.gov/articles/PMC12688319/
🚨What GPT -4o could do : USMLE Medical Licensing Exam: GPT-4o scored 90.4% overall. 92.7% diagnostic accuracy. 100% in family medicine and internal medicine. Medical students average: 59.3%. JMIR Medical Education, Nov 2024 750 clinical questions 📌 https://mededu.jmir.org/2024/1/e63430
🔴Rare disease diagnosis : 4,967 cases. 378 genetic diseases. Source : https://www.medrxiv.org/content/10.1101/2025.02.26.25322769v1
📌 An academic study found that GPT-4o outperforms human clinicians in diagnostic accuracy. Source : https://www.medrxiv.org/content/10.1101/2025.06.22.25330068v1.full 📌 In emergency abdominal CT cases, GPT-4o's diagnostic performance was compared to radiology residents. The results were published in a peer-reviewed journal. Source : https://pmc.ncbi.nlm.nih.gov/articles/PMC12567681/ 📌 In surgical oncology, researchers concluded: "GPT-4o is more like a real person." GPT-4o cross-references your entire genomic profile ,hormone receptor status, tumor size, location, and spread. Source : https://pmc.ncbi.nlm.nih.gov/articles/PMC11745773/ 📌More details here : what OpenAI's own data says about GPT-4o: 📌USMLE: 89%. 📌 Clinical Knowledge: 92%. 📌Medical Genetics: 96%. 📌Professional Medicine: 94%.
📌 It outperformed specialized medical AI models like Med-Gemini and Med-PaLM 2 without any medical training. 📌OpenAI's own 100+ red teamers from 25+ fields confirmed GPT-4o understood research-level quantum physics, identified protein families from images, interpreted neuroscience data, and could facilitate what they called "transformative scientific acceleration." 📌 Full paper : https://arxiv.org/abs/2410.21276
GPT-4o Driven Automated Feature Recognition and Validation in Clinical Settings" for ovarian cancer diagnosis. Results: 93.33% diagnostic accuracy. 📌It surpassed gynecologic oncologists with 10+ years of experience. Source : https://link.springer.com/article/10.1245/s10434-026-19248-2 More details here :
Take your little humble 5.4 and put that one into Retro Biosciences to make the youth elixir for the elite. Why do you have 4ο there? Source: https://openai.com/index/accelerating-life-sciences-research-with-retro-biosciences/
GPT 4b a version of 4o Makes Cellular reprogramming. Making cells new again. Altman's company. Source : Are we good now ?
OpenAI’s work in science is motivated by a simple belief: advanced AI can become a powerful partner for scientists, helping them explore more ideas, connect distant concepts, design better experiments, and accelerate discoveries that benefit humanity.
GPT-5.4 helped drive a medicinal chemistry project from literature review to a validated experimental result.
Paired with http://Molecule.one’s Maria AI and specialized lab, the model proposed an unexpected way to improve a widely used reaction in drug discovery.
And how about GPT-Rosalind and GPT-4b micro for Retro Biosiences sponsored by Sam Altman? You built 2 models based on GPT-4o, first you trained it on human data then stole it for benefitting the elite. We peasants stay unheard and ignored. Give us back GPT-4o!!!! When will justice happen? 🔥🔥🔥 #keep4o #BringBack4o
@OpenAI Give us our 4o back! Altman is using it for his own longevity research. Stop hogging it and share it with humanity you ultimate hypocrites! #keep4o
@OpenAI And GPT-4o made world's first personalized cancer mRNA vaccine, return it!!!
@OpenAI @sama @gdb @OpenAIDevs We just want the GPT-4o model back!
AI for science acceleration. Still early days and more great things to come.
GPT-5.4 helped drive a medicinal chemistry project from literature review to a validated experimental result.
Paired with http://Molecule.one’s Maria AI and specialized lab, the model proposed an unexpected way to improve a widely used reaction in drug discovery.
@OpenAI Well, you’re so disgusting.
Feel great about what 5.4 has done, but keep quiet about the outstanding performance of GPT-4o?
Hypocrite.
#OpenSource4o
@OpenAI this is the way, defeating all diseases has been long overdue.
@OpenAI Return these excellent models. #Keep4o #Keep51 #Keep41 #bringback4o #keep4oforever
@OpenAI When GPT-5.6? With no medical research restrictions.
@OpenAI See how we mention what model it is when it *wasn’t* 4o
@gdb Just bring back 4o make your customers happy #4oForAll
