Z.AI introduced GLM-5.3-Flash on 2026-08-26 as the first natively multimodal model in the GLM-5 series, available through the Z.ai Coding Plan. It is a 320B-total/18B-active MoE that the company says outperforms GLM-5.2 across coding and agentic benchmarks at roughly one-tenth the price, scoring 57 on the Artificial An Architectural changes over GLM-5 include a hybrid sparse + linear attention design to cut long-context serving cost and Manifold-Constrained Hyper-Connections (mHC) for better scaling efficiency, alongside a 30T-token multimodal pre-training corpus. Total parameters sit near GLM-4.5 (320B vs. 355B) but activated parame

Z.AI's official research blog announces GLM-5.2 as the flagship model available through the Z.ai Coding Plan, highlighting a solid 1M-token context that sustains long-horizon coding-agent work, advanced coding with multiple thinking-effort levels, and an MIT open-source license with no regional limits. The model introd On long-horizon coding benchmarks, Z.AI reports GLM-5.2 trails Opus 4.8 by ~1% on FrontierSWE (edging GPT-5.5 by 1% and Opus 4.7 by 11%), ranks second only to Opus 4.8 on PostTrainBench (beating Opus 4.7 and GPT-5.5), and competes on SWE-Marathon ultra-long-horizon tasks — these figures are vendor-reported and not inde

Z.AI released GLM-5.3 on 2026-08-14 as a post-training-only update over the GLM-5.2 base model, accessible via the Z.ai Coding Plan. The company says scaling post-training on its long-horizon environments (IndexShare, SAO, slime stack) produced a 50% gain on its in-house Z.ai Code Bench and open-source SOTA on Terminal An emergent cyber capability is highlighted, with GLM-5.3 scoring 84.5 on CyberGym for vulnerability discovery and more than doubling GLM-5.2 on exploitation benchmarks (ExploitGym 2h/6h of 105/130 and ExploitBench 54.4 vs. 24.4). Open weights are promised about two weeks after launch pending safety evaluation and hard

The provider's catalog lists GLM-5.2 (released 2026-06-13) with text input/output, 1,000,000-token context window and 131,072 max output tokens, plus reasoning, tool calling, structured output, open weights, and temperature controls. Pricing is governed by the Z.AI Coding Plan subscription rather than per-token meterin Social-feed items are useful signals but are third-party and may be stale, so they supplement rather than replace the official Z.ai blog and ZCode documentation. The catalog entry itself serves as a snapshot of the Coding Plan's prior generation model and its integration surface for code-focused workloads.