Meta AI Wins Gold in Five STEM Olympiads, Perfect Physics Score and Corrects Official Answer

Meta officially announced that its internal AI model participated in five international STEM Olympiads under strict conditions with all external tools (such as search, programming, calculators) disabled, achieving gold-medal-level results in all five. It scored perfect in theory exams for both the Asian Physics Olympiad and the International Physics Olympiad, won gold in the International Mathematical Olympiad (top 4% of humans), reached gold level in the Romanian Master of Mathematics, and won gold in the Chemistry Olympiad. Three events were participated in real-time with official scoring. This achievement sparked heated discussion in the community about pure LLM reasoning capabilities.

Confirmed

• Competition results: The model scored perfect and won gold in theory exams for both the Asian Physics Olympiad and the International Physics Olympiad; won gold in the International Mathematical Olympiad (top 4% of humans), reached gold level in the Romanian Master of Mathematics, and won gold in the Chemistry Olympiad. Three events were participated in real-time with official scoring.
• Correcting official answer error: In the International Physics Olympiad (IPhO), the research team found a slight discrepancy between the model's answer and the official preliminary standard answer. After review, the organizing committee confirmed the official answer was wrong, avoiding misjudgment of human contestants, and issued a thank-you letter.
• Test conditions: The test strictly disabled external tools such as search, programming, and calculators to assess uncontaminated true reasoning.

Unconfirmed

• The model's specific parameter count and technical architecture were not disclosed in the posts (@机器之心 mentioned the model is from the MuseSpark series).
• The full name of the specific Chemistry Olympiad event is not clear.

Why it matters

• Validation of pure reasoning: This test effectively verified the AI's true logical level without assistance. Regarding Gary Marcus's view that "pure LLMs cannot solve complex math problems," @inductionheads pointed out that Meta's performance constitutes a strong rebuttal.
• Community reaction: @shuchaabi mentioned netizens humorously joked that researchers who never won gold medals as children can now train a model to win, achieving a "beautiful closed loop."

2026-08-06 ~ 2026-08-07 · 9 related posts

• AI Sweeps Five STEM Olympiads with Perfect Physics Scores — shuchaobi · 2026-08-06
• AI Wins Gold at IPhO and Catches Errors in Official Solutions — shuchaobi · 2026-08-06
• [source] Meta AI Competes in Five STEM Olympiads, Achieves Perfect Physics Scores and Math Gold — AIatMeta · 2026-08-06
• AI Sweeps Five Olympiad Gold Medals, Closing the Loop for Researchers — shuchaobi · 2026-08-07
• Meta's Pure LLM Hits Gold in STEM Olympiads Without Tools, Sparking Debate — inductionheads · 2026-08-07
• Meta Models Achieve Gold-Medal Results in Five International STEM Olympiads — jack_w_rae · 2026-08-07
• Meta's Models Strike Gold in Five STEM Olympiad Competitions — ilkamoi · 2026-08-07

2 near-duplicate retellings: PMinervini · 机器之心