GLM-5.2 ships as 753B-total / 40B-active MoE with 1M context — first open-weight model to beat GPT-5.5 on SWE-Bench Pro, materially compresses the closed-vs-open frontier gap
Zhipu AI's GLM-5.2 ships as a 753B-total-parameter / 40B-active MoE with 1M-token context (5x the 200K GLM-5.1 context). The headline benchmark: first open-weight model to beat GPT-5.5 on SWE-Bench Pro. The closed-vs-open frontier capability gap on production coding workloads is now empirically zero or favorable to open-source for the first time at this benchmark.
The substantive piece is the open-weight-frontier capability inflection on production coding workloads. SWE-Bench Pro is the harder variant of SWE-Bench, designed to be less susceptible to benchmark gaming. GLM-5.2 beating GPT-5.5 on this specific benchmark is meaningful because SWE-Bench Pro is what enterprise-coding procurement actually looks at. The closed-source coding-capability premium that justified API pricing through 2025 is empirically eroded on this benchmark.
The competitive read against Llama 4 Scout's 10M-context and DeepSeek V4 dual-MoE is that the open-weight frontier now has three vendors with distinct capability specializations — Llama 4 Scout for ultra-long-context, DeepSeek V4 for cost-optimized cluster deployment, GLM-5.2 for production coding. The vendor selection optimizes on workload-shape fit, not on 'open vs closed' as a binary choice.
LLM Stats — AI Updates Today (June 2026) – Latest AI Model Releases → · Featherless — Best Open-Source LLMs in 2026 →