GLM-5.2

NIST's Center for AI Standards and Innovation (CAISI) published an independent assessment of Z.ai's GLM-5.2 on July 8, 2026 (posted July 17, 2026), confirming the June 16, 2026 release date and open-weight status. CAISI concludes GLM-5.2 was probably the most capable open-weight model at release, with overall capabilit This assessment provides the strongest third-party capability calibration of GLM-5.2 against named comparators, useful for developers and researchers situating GLM-5.2 on the capability frontier and for policy audiences evaluating open-weight model risk tradeoffs. The cyber-capability and safeguard results are particul

Featherless's blog post reports on Z.ai's GLM-5.2 release on June 16, 2026, describing a ~753B-parameter Mixture-of-Experts model activating ~39B per token, with gains driven by revised long-context architecture and coding-agent training rather than added scale. New architectural elements include IndexShare, which reus On benchmarks, the post reports concrete GLM-5.1 to GLM-5.2 deltas: Terminal-Bench 2.1 rose from 63.5 to 81.0, SWE-bench Pro from 58.4 to 62.1, FrontierSWE from 30.5 to 74.4, SWE-Marathon from 1.0 to 13.0, AIME 2026 from 95.3 to 99.2, and GPQA-Diamond from 86.2 to 91.2. Comparatively, GLM-5.2 is positioned as the highe

A Medium explainer by Mehul Gupta summarizes Zhipu's GLM-5.2 release, framing it as the latest in the GLM family built on GLM-5 and GLM-5.1, with an explicit focus on coding, tool usage, multi-step reasoning, repository analysis, and long-running agent workflows rather than conversational chat. The piece positions GLM- The headline technical highlight reported is a 1 million token context window, presented as enabling AI systems to independently understand, plan, and execute complex development tasks across large, multi-repository projects where traditional context windows are insufficient. The article provides no independent benchma

Z.ai (formerly Zhipu AI) released GLM-5.2 on June 16, 2026, as a 753-billion-parameter open-weights large language model targeted at long-horizon autonomous coding and engineering tasks. The model is available immediately on Hugging Face, the Z.ai API, and more than 20 third-party coding environments, with a stable 1-m Architecturally, GLM-5.2 introduces "IndexShare," which reuses the same indexer across every four sparse attention layers, reducing per-token compute FLOPs by 2.9 times at the maximum 1-million-token context length. The model also features an upgraded Multi-Token Prediction (MTP) layer for speculative decoding, boostin

Z.ai's official developer release notes list GLM-5.2 as a 2026-06-16 entry, highlighting 1M lossless context support for reduced context drift and goal forgetting on long-horizon tasks, open-source SOTA performance on coding and long-horizon task benchmarks, and improved project-level context handling, engineering-stan For developers, the documented GLM-5.2 capability set signals a model targeted at stable behavior on long, complex agentic coding sessions rather than raw scale changes, with the 1M lossless context positioned as a workflow enabler for multi-repository engineering tasks. The release-notes format gives a concise, first-

Databricks published an internal coding-agent benchmark on July 8, 2026, evaluating tools against actual coding tasks their engineers performed on the Databricks multi-million-line codebase, covering Python, Go, TypeScript, Scala, and other popular languages. The benchmark analyzed both task performance and cost across A key conclusion explicitly naming GLM 5.2 was that open models, and GLM 5.2 in particular, are now able to handle even the highest level of task difficulty in their evaluation. The results clustered models and harnesses into three distinct capability tiers, with nuance in which models were effective within each group.