@MaxForAI: A brutal fact: AI's mathematical abilities have already surpassed 99% of humans on this planet. @MenloVentures partner Deedy @deedydas, who invested in Anthropic and OpenRouter, conducted a test. He used the just-concluded 2026 International Mathematical Olympiad...
Summary
AI models Fable, Sol, K3, and Axiom all achieved a perfect score of 42/42 in the 2026 International Mathematical Olympiad, solving the competition completely for the first time at low cost. Among them, Claude Fable 5 was the fastest, while GPT 5.6 Sol had the lowest cost.
View Cached Full Text
Cached at: 07/21/26, 04:46 PM
A brutal fact: AI’s math ability has already surpassed 99% of humans on this planet.
@MenloVentures partner Deedy @deedydas, who has invested in Anthropic and OpenRouter, ran a test.
He used the original problems from the just-concluded 2026 International Mathematical Olympiad (IMO) to test AI.
This is the world’s toughest high school math competition, and it has long been one of the most common benchmarks for evaluating AI models’ mathematical capability.
He put Fable (high), Sol (xhigh), K3 (max), and Axiom through the contest — all scored a perfect 42/42.
For reference, over the past 7 competitions, a total of 4,347 human participants have taken part, and only 30 achieved a perfect score of 42 — a rate of 0.69%.
- Claude Fable 5 solved all problems in a single attempt and was the fastest.
- GPT 5.6 Sol took one extra attempt but had the lowest cost.
- Kimi K3 also managed it, but required 4 extra attempts and consumed a large number of tokens.
Based on the number of attempts and token consumption, problems 3 and 6 were the hardest, followed by problem 2.
Human students have 9 hours to solve these 6 problems; Fable and Sol both finished within 4 hours.
This is the first time the International Mathematical Olympiad has been completely solved.
Last year, the best-performing models were the unreleased Gemini Deep Think and an experimental OpenAI model — both scored 35/42.
This year, three publicly accessible models, including one (K3) about to open-source its weights, accomplished this at a cost of just $10 to $50!
AI’s capability has officially far surpassed the level of the International Mathematical Olympiad.
Deedy (@deedydas): The International Math Olympiad (IMO) 2026, the hardest math contest for high schoolers, just ended.
I ran Fable (high), Sol (xhigh), K3 (max) and Axiom against it and all got a perfect score of 42/42 (repo below if you want to check their solutions): — Claude Fable 5 was the
Similar Articles
@paperpaper886: Last week, I discussed the current state and future of AI4Math with a friend from the math department. He said that current AI is already powerful enough as an auxiliary tool, but there is still a long way to go for AI to achieve independent discovery.
Discussed the current state and future of AI in mathematics. Citing an example, ChatGPT 5.5 Pro autonomously solved the farthest pair problem in high-dimensional computational geometry, which had been stuck for years, demonstrating AI's potential in mathematical discovery.
@VraserX: You genuinely can’t overhype this. GPT-5.6 Sol Ultra, which is a publicly available AI just cracked a 50-year-old math …
GPT-5.6 Sol Ultra, a publicly available AI model, solved a 50-year-old math conjecture in under an hour, suggesting that AI may solve mathematics within the next decade.
@MaxForAI: Tian Yuandong @tydsh's startup team Recursive @Recursive_SI released a milestone: an automated AI research system. In this system, AI can complete the entire research loop of 'propose ideas → implement → run experiments → verify → select next experiment based on results'. Results show that with clear objectives...
The Recursive team released an automated AI research system that can autonomously complete the research loop, surpassing existing human community solutions on multiple benchmarks. For example, on NanoGPT Speedrun it compressed training time from 79.7 seconds to 77.5 seconds, and on SOL-ExecBench it improved the score to 0.754.
@FinanceYF5: 1/ The New Moat of AI Competition: Speed As of 2026/5/30, OpenAI updates major models every 51.8 days on average, Anthropic 59.8 days, Google 75.8 days. The gap is not just in benchmarks, but also in iteration pace.
As of May 30, 2026, OpenAI updates major models every 51.8 days on average, Anthropic 59.8 days, Google 75.8 days, pointing out that AI competition is not only about benchmarks but also about iteration speed.
@FinanceYF5: OpenAI's model just accomplished a major feat: independently solving the plane unit distance problem posed by Erdős in 1946. For 80 years, the best known solution was thought to be a grid-like structure, but AI found a better new construction. This marks the first time AI has independently solved a core open problem in mathematics—a historic breakthrough.
OpenAI's model independently solved the plane unit distance problem posed by Erdős in 1946, marking the first time AI has autonomously solved a core open problem in mathematics—a historic achievement.