Z.AI

Z.AI announced GLM-5.2 on June 16, 2026 as its latest flagship model focused on long-horizon tasks. The release highlights a solid 1M-token context window designed to remain reliable across long, messy coding-agent trajectories rather than just nominally accepting more tokens, with substantially expanded 1M-context tra On the architecture side, GLM-5.2 introduces IndexShare, which reuses the same indexer across every four sparse attention layers to cut per-token FLOPs by 2.9× at 1M context length, and improves the MTP layer for speculative decoding, raising acceptance length by up to 20%. Z.AI positions GLM-5.2 on long-horizon coding

Local AI Zone's September 2026 AI Model Updates roundup covers the month's frontier launches between July 31 and September 11, 2026, noting five launches in ten days (Fable 5.1, GPT-6 Astra, Gemini 3.8 Flash, Muse Spark 1.3, and V4.1-Flash). For Z.AI specifically, the roundup flags GLM-5.3-Flash at a $0.17 per million The dispatch groups Z.AI's GLM-5.3-Flash into broader trends including post-training environment scaling, extreme MoE sparsity, linear attention, and tiered cyber access. It highlights that four of the five frontier launches shipped cyber-capability tiers with gated access programs, with Mercury 2 setting a 1,009 token

On August 14, 2026, Z.ai released GLM-5.3 with substantial gains in long-horizon coding and a notable jump in cybersecurity capabilities — reportedly already used to find a "potentially serious vulnerability in Cursor" according to a Z.ai developer advocate on X, with VentureBeat awaiting Cursor's confirmation. GLM-5.3 The release is significant because GLM-5.3 uses the same 743B-parameter base model as GLM-5.2, with all improvements coming from scaling post-training across more environments, more diverse tasks, and additional RL compute — effectively testing how far a frontier base model can be pushed without another pretraining cyc

A June 17, 2026 DataCamp explainer walks through GLM-5.2 as Z.AI's new flagship in the GLM-5 lineage, highlighting the jump from 200,000 to 1,000,000 tokens via the glm-5.2[1m] identifier, maximum output of 131,072 tokens, and two reasoning effort levels (high and max) with max recommended for complex task stability. T The explainer also covers GLM-5.2 architectural work under the hood, including custom mechanisms designed for long-context agentic workloads, and explains how developers can switch to it inside Claude Code, OpenClaw, and Cline by swapping the Anthropic-compatible base URL and model name. MIT-licensed weights are descri

MarkTechPost reported on Z.AI's GLM-5.2 launch, noting it is the third major release in the GLM-5 line and the fourth flagship-tier coding release in roughly four months (following GLM-5, GLM-5-Turbo, and GLM-5.1). The standout specification is a 1,000,000-token context window — roughly a 5× jump from GLM-5.1's 200,000 The release also adds two thinking-effort levels, High and Max, with Z.AI recommending Max for complex multi-step coding work; in Claude Code the /effort command controls this setting, with xhigh, max, and ultracode options all mapping to GLM-5.2's Max effort. The article frames GLM-5.2 as a long-horizon coding model a

GLM-5.1 Large Model Deeply Empowers Full Flagship Mini PC Lineup Free Exclusive Coding Plan With Every Purchase Recently, GMKtec, a desktop AI computing brand, and Z.AI, an AI large model company, officially announced a major strategic partnership. Z.AI’s newly released flagship model GLM-5.1 is now being integrated ac

Z.AI's official "New Released" documentation index tracks the GLM-5.x line. The June 16, 2026 entry for GLM-5.2 notes support for 1M lossless context that significantly improves long-horizon task capabilities by reducing context drift and goal forgetting in complex tasks, plus open-source SOTA performance on coding and The index also documents subsequent releases: GLM-5.3 on August 18, 2026, which delivers a 50% coding gain over GLM-5.2 on Z.AI Code Bench and reaches open-source SOTA on Terminal Bench 3.0, while matching Mythos 5 in white-box code review and vulnerability discovery and contributing to identification of 2,436 real-wor

This CheapestInference analysis covers GLM-5.3-Flash (API id glm-5.3-flash), Z.ai's efficiency-tier sibling to GLM-5.3. Artificial Analysis measures it at 57 on its Intelligence Index at a blended $0.10 per 1M tokens (~$0.09 per index task), placing it on the intelligence-vs-cost Pareto frontier and ranked 4 of 111 ope Flash weights are published on Hugging Face under plain MIT (zai-org/GLM-5.3-Flash), in contrast to the flagship's custom license carrying a security-review clause for large Model-as-a-Service operators. Specifications include a 1M-token context, up to 128K output, reasoning permanently on, and list pricing of $0.15 in