Researcher Disputes 76% HLE Score Claim for Astra

Florian Brand, a research engineer at Prime Intellect, replied to @teortaxesTex on X that he does not think the HLE benchmark has 76% solvable tasks. The exchange concerns an earlier post that reported high scores for Astra across several AI benchmarks, including FrontierMath and Riemann Bench. Brand's statement directly challenges the 76% figure cited in the discussion of the Astra evaluation.