GPT-6 Astra leaps 4 points in AI benchmark overhaul
artificial analysis updates intelligence index v4.2 after gpt-6 astra scoring drew skepticism. new benchmarks and 40% private test data aim to prevent gaming.
artificial analysis updated its intelligence index to v4.2, driven by skepticism over gpt-6 astra's initial scoring. earlier evaluations had ranked astra highly, but artificial analysis had scored it similarly to its predecessor. the updated index gives astra a four-point gain, placing it second behind anthropic's claude fable 5.1. astra also uses fewer tokens per task than other frontier models. the update includes new benchmarks, such as aa-briefcase and gdp.pdf, and increases private test data weighting to 40%.