GLM-5.2

A July 2026 comparison piece from MarkTechPost evaluates three open-weight MoE models—Moonshot AI's Kimi K3 (2.8T total), DeepSeek V4 Pro (1.6T total), and Zhipu AI's GLM-5.2 (744B total)—on capability, license, and serving cost. While GLM-5.2 is the smallest by total parameter count, the article notes it earned its pl For GLM-5.2 specifically, the piece lists 744B total parameters with 40B active, a 1M-token context with 131K output, and text-only modality, attributed to Zhipu AI (Z.ai) with a June 13, 2026 release date. The article positions GLM-5.2 within the broader landscape of Chinese open-weight models targeting long-horizon c

Lambda's July 9, 2026 blog discusses the reception of GLM-5.2 after its June 16 release by Z.ai, noting that at announcement it advertised scores at or near Anthropic and OpenAI's frontier models and far ahead of GLM-5.1. The piece characterizes the subsequent community reaction as a potential "DeepSeek moment for agen The article highlights anecdotal evidence of enterprise adoption, including a Databricks MTS noting increased corporate requests to serve the model internally, and reports that experienced labs and industry leaders began replacing workloads with GLM-5.2 during extensive testing. The piece frames the open-weight nature—

Z.ai announced GLM-5.2 on June 16, 2026 as its latest flagship model targeting long-horizon tasks. The release marks a substantial leap in long-horizon capability over predecessor GLM-5.1 and is the first time this capability is delivered on a solid 1M-token context. To make that context engineering-usable, Z.ai substa Key technical changes include the new IndexShare architecture, which reuses the same indexer across every four sparse attention layers and reduces per-token FLOPs by 2.9× at a 1M context length. The MTP layer was also improved for speculative decoding, increasing acceptance length by up to 20%. The model ships under a

GLM-5.2 was released as an open-weight model by PRC-based Z.ai (formerly Zhipu AI) on June 16, 2026, according to a July 8, 2026 assessment from NIST's Center for AI Standards and Innovation (CAISI). The report states that GLM-5.2 was probably the most capable open-weight AI model at release, with overall capabilities CAISI's evaluation found GLM-5.2's safeguards and security performance to be mixed: its safeguards allow assistance with agentic cyber exploit development, and it blocks fewer sensitive biological questions than reference U.S. models, but it appears potentially more robust against agent hijacking and prompt-based jailb