OSS
GLM-5.2 Reported as July's Top Open-Source Model: 744B MoE, 91.2% GPQA Diamond, 62.1% SWE-bench Pro
Z.ai's GLM-5.2, a 744-billion-parameter mixture-of-experts model, is being ranked as the strongest open-source LLM available in July 2026, with reported scores of 91.2% on GPQA Diamond and 62.1% on SWE-bench Pro at a fraction of frontier API pricing. Unlike Kimi K3 — announced at 2.8T parameters but with weights not shipping until July 27 — GLM-5.2 is already downloadable and is listed in Ollama's supported-model line. The scores come from aggregator leaderboards rather than a first-party technical report, so treat the exact figures as provisional.
Source
↳ Follow the thread