Stay Ahead of the Curve
Latest AI news, expert analysis, bold opinions, and key trends — delivered to your inbox.
OpenAI Claims Gold-Level Math Olympiad Performance with New Reasoning Model—But Not Everyone’s Convinced
3 min read OpenAI claims its experimental AI just scored gold-level at the Math Olympiad—solving 5 of 6 problems without tools or internet. DeepMind says “not so fast,” arguing the real IMO rules weren’t followed. The kicker? The model isn’t even public yet. July 21, 2025 16:07
OpenAI just announced a major milestone: its experimental general reasoning LLM reportedly achieved gold-level performance in a test modeled after the 2025 International Math Olympiad (IMO)—a benchmark once considered far beyond AI’s reach.
The details:
-
The model tackled two 4.5-hour exams, solving complex problems without any external tools or internet access, just like human contestants.
-
According to OpenAI, it solved 5 out of 6 problems, earning a score of 35 out of 42—enough to win a gold medal at the actual IMO.
-
Submissions were graded independently by three former IMO medalists, and final scores were assigned only through unanimous agreement.
-
But not everyone’s celebrating: Google DeepMind pushed back, arguing that the IMO uses internal marking rubrics, and no external group can officially claim a medal without following them.
Why it matters
Cracking the IMO has long been viewed as one of the holy grails of AI reasoning—a feat that demands not just intelligence, but deep creativity and abstraction. That OpenAI's unnamed experimental model pulled this off marks a huge leap in symbolic and mathematical reasoning.
But the fact that it’s not yet publicly available adds intrigue—and skepticism. If true, OpenAI may be holding even more powerful models behind closed doors, raising fresh questions about transparency, AI benchmarking, and how we define success in machine intelligence.
User Comments (0)
Editor's Choice
AI Agents