IMO
The International Mathematical Olympiad, an international competition used as a significant benchmark for evaluating the mathematical reasoning capabilities of AI systems.
Core Facts
In 2024, Google DeepMind's AlphaProof and AlphaGeometry 2 achieved a silver medal level on IMO-level problems for the first time.
DeepMind's AlphaProof and AlphaGeometry have demonstrated strong capabilities on International Mathematical Olympiad (IMO)-level problems, but are essentially specialized systems.
In 2024, AlphaProof and AlphaGeometry 2 jointly solved 4 out of 6 problems at the IMO competition, equivalent to a silver medal score.
Timeline (last 90 days)
Scale is one of the earliest companies to take IOI (International Olympiad in Informatics) and IMO (International Mathematical Olympiad) backgrounds to the extreme.
In 2024, AlphaProof achieved notable results on International Mathematical Olympiad problems.
Model scores on competition mathematics (such as AIME and IMO) and code generation (such as Codeforces competition problems) have shown a near-vertical growth curve over the past year.
IMO problems cannot be solved directly through memorized training data or statistical patterns but require constructing rigorous logical proof chains, which is considered a more reliable standard for testing whether AI truly understands mathematical structure.
The International Mathematical Olympiad (IMO) was founded in 1959, is held annually, targets high school students worldwide, and covers areas including algebra, geometry, number theory, and combinatorics.
Previously, AI systems had long been stuck at the silver or bronze medal level on IMO problems.
The decarbonization roadmaps of both ICAO (International Civil Aviation Organization) and IMO (International Maritime Organization) list synthetic fuels as a core solution rather than full electrification.
In the second year after 2024, models from multiple institutions were able to solve all IMO problems except the last one.
Before 2024, AI was almost unable to answer IMO-level problems.
In 2024, a model introduced by Google DeepMind answered 4 out of 6 problems correctly on the IMO exam.
20 more timeline events
All Facts (20)
In 2024, Google DeepMind's AlphaProof and AlphaGeometry 2 achieved a silver medal level on IMO-level problems for the first time.
85%VerifiedDeepMind's AlphaProof and AlphaGeometry have demonstrated strong capabilities on International Mathematical Olympiad (IMO)-level problems, but are essentially specialized systems.
80%VerifiedIn 2024, AlphaProof and AlphaGeometry 2 jointly solved 4 out of 6 problems at the IMO competition, equivalent to a silver medal score.
75%VerifiedDeepMind's AlphaProof combined large language models with Lean 4 in 2024 and used reinforcement learning to self-play in a formal environment to generate and verify proof steps.
70%VerifiedDeepMind's AlphaProof and AlphaGeometry, introduced in 2024, together solved 4 out of 6 problems in an IMO test, approaching the gold medal threshold.
65%VerifiedThe International Mathematical Olympiad (IMO) was founded in 1959 and is attended by more than 100 countries and regions each year.
65%UnverifiedAIME and IMO are standard benchmarks for measuring AI mathematical reasoning ability.
70%UnverifiedThere is an essential gap between IMO competition problems and unsolved world-class conjectures, as unsolved conjectures often require inventing entirely new mathematical objects or frameworks.
70%UnverifiedThere is a fundamental gap between solving competition problems and solving Millennium Prize Problems: competition problems have clear solution paths and known answers, whereas Millennium Prize Problems require entirely new mathematical ideas that humanity has yet to discover
60%UnverifiedThe stacking of AI technologies has brought it to a near gold-medal level on International Mathematical Olympiad (IMO)-level problems.
60%UnverifiedAlphaProof is the first AI system in history to reach silver medal level at the IMO.
60%UnverifiedScale is one of the earliest companies to take IOI (International Olympiad in Informatics) and IMO (International Mathematical Olympiad) backgrounds to the extreme.
50%UnverifiedModel scores on competition mathematics (such as AIME and IMO) and code generation (such as Codeforces competition problems) have shown a near-vertical growth curve over the past year.
50%UnverifiedIMO problems cannot be solved directly through memorized training data or statistical patterns but require constructing rigorous logical proof chains, which is considered a more reliable standard for testing whether AI truly understands mathematical structure.
50%UnverifiedPreviously, AI systems had long been stuck at the silver or bronze medal level on IMO problems.
50%UnverifiedThe International Mathematical Olympiad (IMO) was founded in 1959, is held annually, targets high school students worldwide, and covers areas including algebra, geometry, number theory, and combinatorics.
50%UnverifiedThe decarbonization roadmaps of both ICAO (International Civil Aviation Organization) and IMO (International Maritime Organization) list synthetic fuels as a core solution rather than full electrification.
50%UnverifiedThere is a methodological difference between being able to solve a problem with appropriate prompting and reliably solving it under competition conditions, since human contestants have only one attempt.
50%UnverifiedThe IMO problem requires finding the minimum number of tiles needed to satisfy the property and rigorously proving that any other configuration must use more tiles.
50%UnverifiedThe IMO is held once a year; each participating country may send at most 6 contestants; the competition consists of 6 problems answered over two days, with a maximum of 7 points per problem and a total perfect score of 42 points.
50%