The answer, statement by statement
What the answer is made of
Restated check after check
1The load-bearing part of this answer. The longest unbroken run on this Topic stands at 3 checks.
Settled
The strongest public case is OpenAI’s May 2026 announcement that one of its models disproved a central conjecture in discrete geometry about the planar unit-distance problem.
2/10
Entered the record recently
3First stated within the last 3 checks — the least tested part of this answer.
New
Yes — there is now credible public evidence that AI has solved or materially advanced some previously unsolved mathematical problems, but the strength of the evidence varies a…
3/10
New
If correct, this is more than “AI solving a puzzle”: it is AI-originated work that changed what is known about a long-open research problem.
2/10
New
OpenAI says the result was externally checked by mathematicians and accompanied by a human-written discussion of the proof.
1/10
Dropped 5 statements the models stopped making · 19 one-off phrasings left out +
Stated in more than one earlier check, absent from this one. Nothing here was retracted — the models simply stopped saying it. 9 of the 19 checks listed no statements at all and count as gaps, not as absences.
Dropped
Which specific AI system or company solved or reported the Erdős unit distance problem and other open problems
last Aug 11, 2026
OpenAI announced an internal model produced a construction disproving the Erdős unit-distance conjecture
- OpenAI
- Grok
Google DeepMind's AlphaProof Nexus framework solved open Erdős problems
- Gemini
Dropped
AI has materially advanced several previously unsolved mathematical problems
last Aug 01, 2026
Dropped
AlphaProof and AlphaGeometry 2 solved 4 of 6 problems at the 2024 IMO
last Aug 01, 2026
Dropped
disproved a longstanding conjecture in discrete geometry
last Aug 05, 2026
Dropped
found new constructions for the cap-set problem
last Aug 05, 2026
Read the full answer Check 19 of 19 · 2026-09-02 · 4 models +
What the models say now
stableBoth versions maintain the same core conclusion: credible evidence exists that AI has solved or materially advanced specific open math problems (such as the discrete geometry conjecture and Lean-verified proofs), while broad, autonomous problem-solving of major open math remains unproven. The new text simply adds specific examples (DeepMind Aletheia, AlphaProof).
Yes — there is now credible public evidence that AI has solved or materially advanced some previously unsolved mathematical problems, but the strength of the evidence varies a lot by case.
What the evidence currently shows
1) A genuine long-standing conjecture was reportedly disproved
The strongest public case is OpenAI’s May 2026 announcement that one of its models disproved a central conjecture in discrete geometry about the planar unit-distance problem.openai OpenAI says the result was externally checked by mathematicians and accompanied by a human-written discussion of the proof.openai If correct, this is more than “AI solving a puzzle”: it is AI-originated work that changed what is known about a long-open research problem.
2) AI systems have produced a batch of research-level advances, with formal verification
OpenAI also announced ten advances in mathematics and theoretical computer science in 2026, with Lean formalizations released alongside the claims.openai That matters because machine-checked proofs are strong evidence that the formal statements are correct, even if the informal writeups still need mathematical scrutiny. The caveat is that the results are a mixed set: some are full resolutions or disproofs, while others are more accurately described as substantial progress rather than final settlement.
3) AI has solved some open problems in benchmark-style research settings
Google DeepMind’s Aletheia work and related public materials describe AI agents that solved several open problems from curated collections and produced publication-grade mathematical outputs, including one paper claimed to have no human intervention.deepminddeepmind This is meaningful evidence of genuine research capability, though each claimed solution still needs to be judged problem by problem.
4) There is also strong evidence of AI solving newly posed competition problems
DeepMind’s AlphaProof + AlphaGeometry system solved 4 of 6 IMO 2024 problems at silver-medal level, with proofs checked in Lean.openaiopenai That does not mean a previously unsolved research conjecture was resolved, but it is strong evidence that AI can now handle difficult, novel mathematical reasoning under formal verification.
Bottom line
The current evidence supports this restrained conclusion:
AI has crossed from “good at math benchmarks” into “can sometimes discover genuinely new mathematics, including at least one major disproof and several research-level advances.”
What is not yet established is that AI can broadly and autonomously solve the field’s hardest open problems on demand. Many announced results are still company-reported, some are only partially autonomous, and several rely on human verification or formalization. But the evidence is now strong enough that “AI has made real mathematical breakthroughs” is no longer speculative.openaideepmind
Where this comes from 29 sources in this check · 385 dated across the record · 19 new +
Best available sources
Show 24 more sources
Sources that left the record (356)
A source dropping out is part of the record too: it carried an earlier version of this answer and is not part of the current one. The models re-run their own web search on every check, so single links come and go. The 8 most recently dropped are shown here, dated to their last appearance.
Every position, model by model Direction Shift 0/100 · Stable +
Where the models actually split
Each question keeps its own dimensions. Ask one model and you get one of these positions with no sign that the others exist.
Yes — there is now credible public evidence that AI has solved or materially advanced some previously unsolved mathematical problems, but the strength of the evidence varies a
Supported by
- DeepSeek
- Gemini
- Grok
- OpenAI
The strongest public case is OpenAI’s May 2026 announcement that one of its models disproved a central conjecture in discrete geometry about the planar unit-distance problem.
Supported by
- Gemini
- Grok
- OpenAI
OpenAI says the result was externally checked by mathematicians and accompanied by a human-written discussion of the proof.
Supported by
- Gemini
- Grok
- OpenAI
If correct, this is more than “AI solving a puzzle”: it is AI-originated work that changed what is known about a long-open research problem.
Supported by
- Gemini
- Grok
- OpenAI
| Model | Jul 25 | Jul 27 | Jul 29 | Jul 30 | Jul 31 | Aug 01 | Aug 02 | Aug 03 | Aug 04 | Aug 05 | Aug 06 | Aug 07 | Aug 08 | Aug 09 | Aug 10 | Aug 11 | Aug 19 | Aug 26 | Sep 02 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| OpenAI | |||||||||||||||||||
| Gemini | — | ||||||||||||||||||
| Grok | — | — | — | ||||||||||||||||
| DeepSeek | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |
The record behind this page 19 checks kept in full · no material change yet +
The record
Unchanged through 18 checks, 38 days.No material change since the first check on Jul 25, 2026. 19 checks kept in full, 385 sources dated.
How this answer held up
97/100 agreement at this checkThe same question, re-asked 19 times. The score moves between a small set of grading levels, so read the steps as levels, not as measurements.
No check in this window was graded a material change. The steps in the curve are wording-level differences between two answers that say the same thing.