AI Models Solving Advanced Math Problems
On March 23rd, researchers at Epoch AI, alongside the original author of the problem, confirmed something that would have seemed faintly absurd two years ago: GPT-5.4 Pro had solved an open problem from the FrontierMath benchmark — a set of problems specifically designed to be hard enough that automated systems weren't supposed to touch them yet. The claim earned a singularity score of 8.4, suggesting the observers considered it genuinely discontinuous from prior performance. Then, quietly, the story escalated. By April 1st — a date that will not go unnoticed — an OpenAI internal model reportedly solved three more problems from the Erdős collection, a set of conjectures that have resisted professional mathematicians for decades. No public model name was attached to that second claim, which is either reassuringly humble or conspicuously convenient depending on your disposition.
If confirmed, here is what this means. The FrontierMath benchmark was constructed precisely to measure the gap between human mathematical reasoning and machine pattern-matching — and if that gap is closing on open problems, not textbook exercises, the implications run far beyond competitive benchmarks. Professional mathematicians working on combinatorics, number theory, and discrete mathematics would find themselves with an AI collaborator capable of genuine contribution, not just verification. For OpenAI, it would represent a meaningful shift in how the company positions its research models — less consumer product, more scientific instrument. The second-order effect is subtler but worth naming: if internal models are solving Erdős problems before public release, we are in a period where the frontier of AI capability is meaningfully ahead of what any external researcher can observe or audit. That asymmetry matters to everyone trying to understand where this technology actually is.
Watch for independent verification from the Erdős problem community or FrontierMath's own leaderboard updates — and pay close attention to whether OpenAI formalises the April 1st claim with a named model and reproducible methodology.
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1x5y9ue/gpt6_astra_autonomously_solves_49_out_of_50/"> <img alt="GPT-6 Astra autonomously solves 49 out of
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1wyxdtu/for_31_years_mathematicians_couldnt_finish_this/"> <img alt="For 31 years, mathematicians couldn’t
<table> <tr><td> <a href="https://www.reddit.com/r/OpenAI/comments/1waylo4/openai_llms_has_resolved_a_roughly_90_years_old/"> <img alt="OpenAI LLMs has resolved a roughly 90 years
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1s3pbl6/human_vs_ai_performance_on_arcagi_3_as_a_function/"> <img alt="Human vs. AI performance on ARC-AGI
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1v3mxrc/the_dinitzgarggoemans_conjecture_is_false/"> <img alt="The Dinitz-Garg-Goemans conjecture is false"
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1tgn8qt/gemini_32_flash_is_capable_of_solving_imo_2025_p6/"> <img alt="Gemini 3.2 Flash is capable of
<!-- SC_OFF --><div class="md"><p>So many things happened in just 7 days:</p> <ul> <li>OpenAI rolled out GPT-6 to more than 1.2 billion ChatGPT users. Answers now come with
<!-- SC_OFF --><div class="md"><p>On September 10, mathematicians published an open letter demanding the cancellation of the Caltech Mathathon, a student-run hackathon where 100
NewsHive monitors these sources continuously. All signal titles above link to the original reporting.
Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →