Jasper Dekoninck Shares GPT-6-Astra MathArena Lead
PhD student at ETH Zurich highlights model's top ranking and efficiency on benchmarks.
TLDR
A tweet from Jasper Dekoninck, a PhD student at ETH Zurich focused on LLM math evaluations, states that GPT-6-Astra holds first place on the MathArena leaderboard with 90% expected performance. The post adds that the model is token efficient, averaging 12k tokens per question on BrokenArXiv and ArXivMath. It includes a screenshot of the MathArena leaderboard UI showing the ArXivLean tab for 06/2026 along with tabs for BrokenArXiv, ArXivMath, and others.
Combined views
—
1 Source, first seen ago
