What AI can now do·Jul 19, 2025·Math
AI reaches IMO gold: 35/42 under contest rules, from OpenAI and Google
35/42 (gold)IMO 2025 score
An experimental OpenAI model and Google's Gemini Deep Think each solved 5 of 6 IMO 2025 problems for 35/42 in natural language within the two 4.5-hour sessions; a year earlier AlphaProof managed silver (28/42) on formalized problems with up to three days per problem.
How far it has travelled
Demonstratedshown once, by its developer or a researcher
yes · Jul 19, 2025 · OpenAI (Jul 19): 35/42, no tools, graded by three former IMO medalists; DeepMind (Jul 21): 35/42, graded by IMO coordinators, who confirmed the proofs but did not validate the system. source
Reproducedshown again by someone else
yes · Jul 21, 2025 · Two labs reached the same score independently within three days; AlphaProof and AlphaGeometry 2 had scored 28/42 (silver) in 2024 with manually formalized problems. source
Accessibleavailable to the public or to customers
partial · Aug 1, 2025 · A faster Deep Think variant (bronze-level on IMO 2025 internally) went to Google AI Ultra subscribers on Aug 1, 2025; OpenAI said it would not release that level of math capability 'for several months'. source
Reliableworks most of the time on a defined task
partial · May 31, 2026 · Later systems resolved Erdos problems checked in Lean or by Terence Tao (Jan-Jul 2026), but an Oct 2025 OpenAI claim of ten Erdos solutions collapsed when the answers were found in the literature. source
Real consequenceshas changed something in the world, for good or ill
yes · Jul 20, 2026 · AI-assisted mathematics produced the first counterexample to the 1939 Jacobian conjecture (Jul 2026) and a run of Erdos problem resolutions; Tao cautions that most are 'low-hanging fruit'. source
What this could change for you
Competition-level math is no longer a human-only skill, which matters most to students, teachers and anyone whose job was being the smartest person in the room.
Sources
primary source, second source.
A milestone is a dated, sourced result that moved what AI can do; a stage is upgraded only when a source shows it (a reproduction, a public release, production evidence, a documented consequence). “Unknown” is the honest default.
