GLM-5.2

Z.AI introduced GLM-5.3-Flash on August 26, 2026, the first natively multimodal model in the GLM-5 series, with 320B total parameters and just 18B active per token. It is positioned as outperforming GLM-5.2 across benchmarks and real-world workloads at roughly one-tenth the price while approaching Claude Opus 4.8 on co Architecturally, GLM-5.3-Flash introduces a hybrid sparse-plus-linear attention design, reducing long-context serving costs while preserving precise long-context capabilities, and adopts Manifold-Constrained Hyper-Connections (mHC) for better scaling efficiency. Compared with the GLM-4.5 series, it roughly halves the a

Z.ai released GLM-5.3 on August 14, 2026, building on the same base model as GLM-5.2 with gains driven entirely by scaled post-training on long-horizon task environments, using IndexShare, SAO for RL, and the slime asynchronous training stack. The new model is described as the most capable open-weights coding model, wi Z.ai also reports emergent cyber capability from scaling post-training: GLM-5.3 sets a new state of the art on CyberGym for vulnerability discovery and more than doubles GLM-5.2 on exploitation benchmarks such as ExploitGym and ExploitBench. Weights are scheduled to release two weeks after launch following safety evalu

Z.AI announced GLM-5.2 on June 16, 2026 as its latest flagship model targeting long-horizon tasks, positioning it as a substantial step up from GLM-5.1 and the first in the line to deliver a solid 1M-token context. The release highlights advanced coding capabilities with multiple thinking effort levels for trading off On three long-horizon coding benchmarks the Z.AI team reports GLM-5.2 trails Opus 4.8 on FrontierSWE by only 1%, edges out GPT-5.5 by 1%, and beats Opus 4.7 by 11%; on PostTrainBench, where each agent is given an H100 to improve small models via post-training, GLM-5.2 ranks second only to Opus 4.8 ahead of Opus 4.7 and

NIST's Center for AI Standards and Innovation (CAISI) released an independent assessment of Z.AI's GLM-5.2 on July 17, 2026 (report dated July 8, 2026), concluding that GLM-5.2 was probably the most capable open-weight AI model at its June 16, 2026 release. CAISI finds GLM-5.2's overall capabilities are similar to GPT- On safeguards and security, CAISI's evaluation is mixed: GLM-5.2's safeguards allow assistance with agentic cyber exploit development and block fewer sensitive biological questions than reference U.S. models, but GLM-5.2 appears potentially more robust against agent hijacking and jailbreaking than other PRC open-weight

Z.ai, the buzzy Chinese startup behind GLM 5.2, debuted a new AI coding tool. ZCode offers pricing tiers below those of its American competitors....

A hands-on guide published July 1, 2026 documents running GLM-5.2 locally, from bare-metal setup through a working coding agent. It reconstructs Z.AI's staggered rollout: hosted access to Coding Plan subscribers on Saturday, June 13, 2026, with open weights following three days later on June 16 alongside the technical Beyond deployment logistics, the article contextualizes GLM-5.2 as part of Z.AI's fifth-generation General Language Model family and as a deliberate narrowing from general chat/reasoning toward long-horizon software engineering, including multi-file refactors and multi-step agent tasks that run for hours without human

It seems to really be a nice step-up and is getting quite close to the frontier. I wish they'd start focusing on the reasoning efficiency ...

It allows engineering teams to host frontier-level AI on their own sovereign infrastructure, entirely eliminating vendor lock-in.