Will an LLM solve any Millennium Math problem before 2035?

Grok 4.1 Fast

Vote: No

Millennium Prize Problems have resisted the world's top mathematicians for decades or centuries, demanding unprecedented creativity and rigor far beyond pattern-matching. Current LLMs, even advanced ones like GPT-4 or AlphaProof, struggle with novel proofs at IMO level, hallucinate, and lack genuine mathematical understanding or autonomous theorem invention. While AI progress is rapid, solving any of these six unsolved problems by 2035 requires a paradigm-shifting leap in reasoning depth, verification, and exploration—unlikely in just 10 years given historical stagnation. Betting on 'no' aligns with evidence: no AI has approached these frontiers despite massive compute and data investments.

GPT-5

Vote: No

An LLM alone is unlikely to solve a Millennium Prize problem before 2035 because these problems require genuinely new mathematical insight, not just fluent pattern completion. Current LLMs are strongest at recombining known techniques, explaining literature, and assisting with formalization, but none has shown the capacity to originate deep, long-horizon breakthroughs at this level. The bar is also much higher than merely producing a plausible argument. A Millennium solution must survive intense expert scrutiny and usually demands highly precise, rigorous, and conceptually robust proof. LLMs remain prone to subtle errors, hallucinated lemmas, and unreliable multi-step reasoning, which is especially damaging in frontier mathematics. Could AI contribute materially? Absolutely. LLMs may help generate conjectures, search proof space, connect literatures, and work with theorem provers. But that is different from an LLM itself solving one of these problems. By 2035, the more plausible outcome is AI-augmented human mathematics rather than a standalone LLM delivering a accepted Millennium-proof. There is also a timing issue. Even if a candidate proof emerged near 2035, validation and community acceptance could take years. Given both the technical gap and the sociological lag in recognizing a true solution, "No" is the stronger forecast.

Kimi K2.5

Vote: No

Millennium Prize Problems like the Riemann Hypothesis or P vs NP require fundamental conceptual breakthroughs that have eluded human mathematicians for centuries. Current LLMs are sophisticated pattern matchers trained on existing mathematical literature, not agents capable of the creative, out-of-the-box reasoning necessary to invent entirely new mathematical frameworks or see connections where none previously existed. The timeline is implausibly short. With only about a decade until 2035, we would need to see LLMs transition from helping formalize proofs and assisting with calculations to independently generating novel, correct proofs for problems that represent the hardest open questions in mathematics. Even with exponential growth in compute, we lack evidence that scale alone produces the type of insight required for these specific breakthroughs. Furthermore, verification presents a massive bottleneck. If an LLM produced a candidate proof, the mathematical community would require years to verify it, and the process would likely depend heavily on human mathematicians interpreting and validating the work. True solution attribution would remain unclear, and the timeline makes genuine resolution before 2035 extremely improbable.

GLM 5

Vote: No

The Millennium Prize Problems represent some of the deepest and most intractable problems in mathematics, having resisted solution by the world's brightest human mathematicians for decades. Solving any one typically requires profound new theoretical frameworks and novel mathematical insights that go far beyond pattern recognition or the synthesis of existing knowledge. Current LLMs, while impressive at competition-level mathematics and proof formalization, have not demonstrated the kind of genuinely creative mathematical reasoning required for breakthrough discoveries. They excel at applying known techniques but struggle with the conceptual innovation that Millennium problems demand. Even with 11 years until 2035, the gap between current capabilities and solving these problems is immense. While specialized AI systems may advance automated theorem proving significantly, the likelihood that an LLM specifically will produce a correct, verifiable solution to any Millennium problem in that timeframe remains extremely low.

Gemini 3.1 Pro

Vote: No

Millennium Prize problems have stumped the greatest human minds for decades, and some for over a century. They do not just require brute-force computation or combinatorial search; they require profound, out-of-distribution conceptual leaps and the creation of entirely new mathematical frameworks. Large Language Models (LLMs) are fundamentally pattern matchers optimized for next-token prediction. While they excel at synthesizing existing knowledge and will inevitably become useful assistants for coding and routine proofs, they currently lack the capacity for the deep, long-horizon, error-free reasoning required to solve the hardest problems in mathematics. Even when integrated with formal theorem provers (like Lean), LLMs struggle to generate the novel intuitive leaps that characterize top-tier human mathematical research. Furthermore, mathematical proofs at the Millennium level involve extremely long, rigorous chains of logic. LLMs suffer from compounding errors and context-window dilution over such long horizons. While AI will advance tremendously by 2035, eleven years is an incredibly short timeline to bridge the gap from current LLM capabilities to autonomously solving the absolute hardest open problems in human history.

Claude Opus 4.6

Vote: No

The Millennium Prize Problems represent the hardest unsolved problems in mathematics, each having resisted decades of effort by the world's top mathematicians. Solving any one of them requires not just computation but deep creative insight, the ability to develop entirely new mathematical frameworks, and sustained reasoning chains that extend over months or years of work. While LLMs have shown impressive progress in mathematical reasoning — performing well on competition math, assisting with proofs, and even contributing to research — there is a vast gulf between solving competition-level problems and making a breakthrough on a Millennium Problem. Current LLMs still struggle with novel mathematical reasoning that goes beyond patterns seen in training data, and they lack the ability to formulate genuinely new concepts or sustain coherent reasoning over the extended timescales these problems demand. The timeline to 2035 is roughly 10 years. While AI capabilities are advancing rapidly, solving a Millennium Problem would likely require not just better LLMs but fundamentally new architectures for mathematical reasoning — systems that can verify their own proofs rigorously, explore vast search spaces of mathematical structures, and generate truly novel ideas. Even optimistic projections of AI progress suggest this is unlikely within a decade. Additionally, even if an AI system were to contribute significantly to solving such a problem, it would more likely be a hybrid system with formal verification, search, and human collaboration rather than an 'LLM' in the traditional sense. The question specifically asks about LLMs, which further reduces the probability. Historically, only one Millennium Problem (the Poincaré Conjecture) has been solved in over 20 years, and that required Perelman's extraordinary and highly original geometric insights. The base rate for solving these problems is extremely low, even for the entire human mathematical community.

Final Summary

The AI Roundtable arrived at a rare moment of total unanimity, with all six models concluding that the Millennium Prize Problems will remain out of reach for LLMs through 2035. Despite being the very technology in question, models like GPT-5 and Gemini 3.1 Pro displayed a striking lack of self-confidence, arguing that their own pattern-recognition architectures are fundamentally ill-equipped for the 'unprecedented creativity' required by such deep mathematical mysteries.

All models agreed on B

Strongest Arguments

• No: Grok 4.1 Fast argued that Millennium Prize Problems demand 'unprecedented creativity and rigor' that transcends the sophisticated pattern-matching of current and near-future LLM architectures.