AI Evaluation Revolution! Artificial Analysis Introduces Industry Capability Index: Claude Fable 5 Dominates, Open Source GLM-5.2 Nearing Closed-Source Cutting Edge

According to Inference Watch monitoring, traditional AI model evaluations often rely on general academic datasets, leading to a discrepancy between high test scores and actual enterprise performance. Evaluation agency Artificial Analysis, building on existing general skills indices for intelligent agents and programming, has introduced six industry-specific indices to assess real-world business performance (covering the fields of finance, law, healthcare, operations, engineering, and economics).

Unlike general skills indices, the industry indices are custom-weighted based on the O*NET Job Activity Classification, simulating the task distribution frequency of real positions. They combine testing of the underlying capabilities of large models (such as HLE reasoning, LCR long-text processing, GDPval agent work, etc.) with each industry's professional knowledge base (AA-Omniscience). In the initial assessment, the closed-source model Claude Fable 5 claimed the top spot in all eight indices, while the open-source GLM-5.2 secured five first places in the six industry indices, scoring 53 in the engineering index, approaching Claude Sonnet 5's 55.

The evaluation also revealed significant efficiency gaps: in the operations index, Claude Fable 5's single-task cost ($3.48) was over a hundred times that of the open-source DeepSeek V4 Pro ($0.03); and in the law index, the Gemini 3.1 Pro preview version was nearly seven times faster than Claude Fable 5 and scored only 11 points lower.

Weighing efficiency against cost-effectiveness based on task scenarios is becoming a new benchmark for AI selection in vertical domains.