GPT-5 is coming soon! OpenAI claims the 2025 IMO Gold Medal.

• OpenAI announced that its latest experimental reasoning large language model (LLM) has achieved gold medal-level performance in the 2025 International Mathematics Olympiad (IMO), solving 5 out of 6 problems with a total score of 35/42.
• The model was evaluated under the same rules as human contestants: two 4.5-hour exam sessions, no tools or internet access, reading official problem statements, and writing natural language proofs. Scores were determined by three former IMO medalists through unanimous consensus.

Significance

• Enhanced Reasoning Capability: IMO problems require a higher level of sustained creative thinking compared to previous benchmarks. The progression in reasoning time horizon has advanced from GSM8K (~0.1 minutes for top humans) → MATH benchmark (~1 minute) → AIME (~10 minutes) → IMO (~100 minutes).
• Breakthrough in Verification: IMO submissions involve hard-to-verify, multi-page proofs. This progress goes beyond the reinforcement learning paradigm with clear, verifiable rewards, enabling the model to generate intricate and rigorous arguments at the level of human mathematicians.

Technical Approach

• The capability was achieved not through narrow, task-specific methodologies but by making breakthroughs in general-purpose reinforcement learning and test-time compute scaling.

Related Plans

• GPT-5 will be released soon, but the IMO gold medal-winning model is an experimental research model. OpenAI does not plan to release products with such mathematical capabilities for several months.
• The model's solutions to the 2025 IMO problems (excluding Problem 6) have been published on relevant platforms.

source from https://x.com/alexwei_/status/1946477762023412135