The answer, statement by statement
What the answer is made of
Entered the record recently
4First stated within the last 3 checks — the least tested part of this answer.
New
Model C claims OpenAI solved ten long-standing open problems and non-sofic groups in August 2026, while the consensus and other models only…
0/6
OpenAI model solved ten major problems including non-sofic groups and Connes' rigidity conjecture in August 2026.
- Gemini
Only the Erdős unit-distance conjecture disproof is discussed as OpenAI's major open problem success.
- Grok
- OpenAI
New
disproved long-standing conjectures or produced machine-checked proofs of open results
1/6
New
formal verification greatly reduces the risk of “looks right but is wrong” AI output
1/6
New
The scale and scope of open problems solved by AI beyond the unit-distance conjecture
0/6
AI systems like OpenAI's Astra have autonomously resolved or made headway on ten major open problems across multiple fields with zero unproven steps.
- Gemini
Only a single prominent breakthrough (the unit-distance conjecture) is clearly documented, while others are collaborative or incremental.
- OpenAI
Dropped 2 statements the models stopped making · 12 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 Nothing here was retracted — the models simply stopped saying it. 3 of the 9 checks listed no statements at all and count as gaps, not as absences.
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
Read the full answer Check 9 of 19 · 2026-08-04 · 3 models +
What the models said then
stableThe consensus answers are extremely similar in core conclusions, structure, and factual claims. Both agree that AI has materially advanced unsolved math, achieved competition-level milestones (IMO silver/gold), and has at least one major documented success (disproving an Erdős/planar unit-distance conjecture), while not solving major Millennium Prize problems. Minor wording and restructuring diffe
As of now, the best evidence suggests that AI has materially advanced mathematical research and has at least one strong publicly documented case of resolving a long-standing open problem, but it has not been shown to have broadly “solved mathematics” or routinely cracked the field’s hardest landmark conjectures.
What the evidence is strongest on
1) Competition-level math: clearly yes
AI systems have reached or exceeded elite human performance on very hard contest problems, including IMO problems.
- DeepMind reported a system reaching silver-medal level on the 2024 IMO, with formally verified solutions in Lean.openai
- DeepMind later reported a gold-medal level result on IMO 2025 with an advanced Gemini “Deep Think” system.openai
This is important evidence of strong reasoning, but IMO problems are still posed problems, not open research questions.
2) Open research problems: there is now serious evidence of real progress
The most compelling publicly documented case is OpenAI’s reported disproof of the planar unit-distance conjecture in discrete geometry.openai
Why this stands out:
- it concerns a genuine, long-standing open problem;
- OpenAI released a proof document;
- external mathematicians publicly commented on it, including prominent experts;
- the result was presented as a genuine mathematical construction, not just a benchmark answer.openai
This is the clearest evidence so far that AI has done more than generate plausible-looking proofs: it appears to have produced a mathematically substantive new result.
3) AI-assisted advances beyond one headline result
There is also substantial evidence of AI helping produce or accelerate research-level results:
- OpenAI described GPT-5 assisting in work on a 40-year-old optimization problem, where the human researcher used the model to help develop the proof strategy.openai
- OpenAI’s “early experiments in accelerating science with GPT-5” describe multiple mathematical and theoretical results where model-generated arguments were checked and incorporated by humans.openai
- DeepMind reported semi-autonomous work on a large set of Erdős problems, with some cases turning out to be genuinely novel and others already solved in the literature, which is itself revealing about the current state of the art.deepmind
These are best understood as material acceleration of research, not full replacement of mathematicians.
What the evidence does not yet show
The current evidence does not support the claim that AI has broadly solved the deepest open problems in mathematics, such as the Millennium Prize Problems. Nor does it show that AI can now routinely produce publishable, end-to-end mathematical research without substantial human oversight, problem selection, or verification.
A careful summary is:
AI has moved from “solving hard exam problems” to occasionally producing genuine new mathematics and materially assisting research, but the field is still in an early phase, and most results remain either collaboration-heavy, lower-bound improvements, or single prominent breakthroughs rather than a general solution to open mathematics.
Bottom line
If you want the most defensible current answer: yes, AI has materially advanced previously unsolved mathematics, and there is at least one highly credible, publicly documented case of an AI-generated solution to a long-standing open problem.openai But the evidence is still selective and emerging, not proof that AI has broadly conquered unsolved mathematics.
Where this comes from 35 sources in this check · 216 dated across the record · 20 new +
Best available sources
Show 30 more sources
Sources that left the record (181)
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.
The scale and scope of open problems solved by AI beyond the unit-distance conjecture
AI systems like OpenAI's Astra have autonomously resolved or made headway on ten major open problems across multiple fields with zero unproven steps.
- Gemini
Only a single prominent breakthrough (the unit-distance conjecture) is clearly documented, while others are collaborative or incremental.
- OpenAI
| Model | Jul 25 | Jul 27 | Jul 29 | Jul 30 | Jul 31 | Aug 01 | Aug 02 | Aug 03 | Aug 04 |
|---|---|---|---|---|---|---|---|---|---|
| OpenAI | |||||||||
| Gemini | |||||||||
| Grok | — | — |
The record behind this page 9 checks kept in full · no material change yet +
The record
Unchanged through 8 checks, 10 days.No material change since the first check on Jul 25, 2026. 9 checks kept in full, 216 sources dated.
How this answer held up
64/100 agreement at this checkThe same question, re-asked 9 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.