Open-source GLM-5.2 is ridiculously good value—its cost is less than one quarter of Opus 4.8, yet it’s only 90 points behind. After reading this review, I couldn’t help but think, “So worth it.”

Artificial Analysis launches its first long-term knowledge work evaluation benchmark for large model intelligent agents, covering four scenarios: data science, product management, banking operations, and heavy industry strategy, with 91 tasks developed by experts from Google, McKinsey, and Boston Consulting. The results show that Claude Fable 5 took the top spot, followed by Opus 4.8, with GLM-5.2 ranking in the top three; however, under the all-correct standard for individual items, Fable 5's perfect rate is only 3%. The open-source GLM-5.2 has a composite score only 90 points lower than Opus 4.8, but costs less than a quarter of it.