GLM-5.1

The Morph LLM technical digest (updated July 2026) consolidates Z.ai's GLM-5.1 specifications: a 744B total / 40B active MoE model with a 200K-token context window, up to 128K-131K max output, and DSA sparse attention, released April 7, 2026 under the MIT license on Hugging Face. It scored 58.4 on SWE-bench Pro at laun The same source documents that GLM-5.1 was a post-training upgrade to GLM-5 with roughly 28% coding improvement from refined post-training and asynchronous agent RL, trained to sustain up to 8 hours of autonomous execution across hundreds of tool-call rounds. Runtime support includes transformers, vLLM, SGLang, KTransf

Mehul Gupta's Medium walkthrough positions GLM-5.2 (released in June 2026, just months after GLM-5.1) as Zhipu's doubled-down bet on agentic software engineering, with a 1 million token context window, new reasoning modes tuned for coding, and explicit emphasis on tool usage, multi-step reasoning, repository analysis, For developers coming from the Wafer-hosted GLM-5.1, the piece traces the same release lineage — GLM-5 → GLM-5.1 → GLM-5.2 — and explains how the 1M-token context window and coding-focused reasoning modes inherited from and expanded upon in GLM-5.1 underpin GLM-5.2's open-weight coding performance, offering a migration

For developers evaluating Wafer's GLM-5.1 against other hosted open-weight options, Artificial Analysis provides useful quantitative baseline figures for the same model architecture: GLM-5.1 sits at an Intelligence Index of 40, uses roughly 26k output tokens per Intelligence Index task, and costs about $0.25 per task o The same Artificial Analysis article confirms that GLM-5.2 (the immediate successor) is priced at $1.4 per 1M input, $4.4 per 1M output, and $0.26 per 1M cache-hit tokens on Z.ai's first-party API, and is MIT-licensed. While this is GLM-5.2 data rather than GLM-5.1-specific, it gives Wafer customers a reference point f

Z.ai's GLM series sits within the Tier 4 Chinese frontier models grouping in the May 9, 2026 LLM Encyclopedia, alongside ERNIE, Kimi, Qwen, Hunyuan and others, reflecting GLM-5.1's positioning as an open-weight coding-focused model family competing with proprietary leaders on benchmarks and developer tools. Although the explainx.ai blog is primarily framed around the GLM-5.2 release and the so-called Fable 5 export-ban episode, it provides concrete ecosystem-level signal relevant to GLM-5.1's current standing: GLM-5.2 remains text-only while topping open coding benchmarks, Kimi K2.7 Code is now GA in GitHub Copilot, and C

The Stochastic Sandbox LLM Encyclopedia (May 9, 2026) catalog of 60+ models from 22 use cases places Zhipu's GLM-5 and GLM-5.1 (alongside ChatGLM) under Tier 4 — Chinese Frontier Models, alongside Baidu ERNIE, Moonshot Kimi, Baichuan, 01.AI Yi, Hunyuan, InternLM and ByteDance Seed, as pricing and benchmark tables are r For developers evaluating the Wafer-hosted GLM-5.1 endpoint, this encyclopedia serves as an independent third-party reference confirming GLM-5.1's place in the open-weight Chinese frontier tier; however, the supplied excerpt does not contain GLM-5.1-specific benchmark deltas, pricing, or Wafer routing details, so it fu

Z.ai releases GLM-5.1, an open-source model scoring 94.6% of Claude Opus 4.6 in coding benchmarks. Full breakdown of benchmarks, pricing, architecture, and what's still unverified.

The Chinese company said its new open-source model can continue to improve over hundreds of iterations, as AI vendors race to build tools that can handle longer software tasks.