GLM-5.2 has achieved notable recognition by placing third in the GDPval-AA benchmark for real-world economic knowledge tasks, surpassing GPT-5.5 and earning the distinction of the leading open weights model. The model, developed by Zai_org, scored 1524 Elo on the GDPval-AA, which measures performance through long-horizon, multi-turn tasks that simulate meaningful economic work. This demonstrates a trend in which open weights models are becoming increasingly competitive against proprietary systems, highlighting the growing accessibility of advanced AI capabilities and the shift toward benchmarks that emphasize practical utility in real-world applications.

GLM-5.2: GLM-5.2 is an advanced AI model released by Zai_org. It ranks third overall on the GDPval-AA benchmark for real-world agentic knowledge work while leading all open weights models by a significant margin. The model also tops open weights performance on the Artificial Analysis Intelligence Index, Agentic Index, and AA-Briefcase.
GPT-5.5: GPT-5.5 is a proprietary AI model that achieves results comparable to GLM-5.2 on the GDPval-AA benchmark. It trails the top entries in this real-world evaluation.
Zai_org: Zai_org develops the GLM family of AI models. It launched GLM-5.2, which demonstrated competitive results against leading proprietary systems on multiple evaluation frameworks focused on practical tasks. The organization emphasizes broad accessibility for its releases.
GDPval-AA: GDPval-AA is a benchmark from Artificial Analysis that assesses AI models on real-world, economically valuable knowledge work using multi-turn agentic tasks. GLM-5.2 achieved strong results on it, averaging around 31 turns per task across nearly two thousand matches.
MiniMax-M3: MiniMax-M3 is an open weights AI model that ranks as the next strongest after GLM-5.2 on the GDPval-AA benchmark.
Muse Spark: Muse Spark is a proprietary AI model evaluated on the GDPval-AA benchmark where it trails GLM-5.2.
AA-Briefcase: AA-Briefcase is a benchmark from Artificial Analysis focused on AI model capabilities. GLM-5.2 ranks third on this evaluation.
Qwen 3.7 Max: Qwen 3.7 Max is a proprietary AI model evaluated on the GDPval-AA benchmark where it trails GLM-5.2.
Agentic Index: Agentic Index is an evaluation framework from Artificial Analysis that measures AI model performance on agentic tasks. GLM-5.2 ranks third on this index.
Claude Fable 5: Claude Fable 5 is a proprietary AI model that leads the GDPval-AA benchmark. It outperforms GLM-5.2 on this real-world agentic evaluation.
Claude Opus 4.8: Claude Opus 4.8 is a proprietary AI model that ranks second on the GDPval-AA benchmark. It sits ahead of GLM-5.2 in overall scoring.
Artificial Analysis: Artificial Analysis runs comprehensive AI model evaluations including the GDPval-AA benchmark and related indices. It positions GLM-5.2 at number three overall and first among open weights entries based on long-horizon agentic performance. The platform also maintains the Artificial Analysis Intelligence Index, Agentic Index, and AA-Briefcase.
Google’s Gemini 3.5 Flash: Google’s Gemini 3.5 Flash is a proprietary AI model evaluated on the GDPval-AA benchmark where it trails GLM-5.2.

{“Accessibility”: “Leading AI models are increasingly released with broad availability, expanding access to advanced capabilities.”, “Benchmark Trends”: “Evaluations emphasizing long-horizon multi-turn agentic work are becoming central to assessing practical model utility.”, “Open Weights Performance”: “The GLM-5.2, developed by Zai_org, leads open weights models, showing strong competitive results against proprietary systems on GDPval-AA, an economic agentic benchmark.”}