ZCode review: Z.ai's GLM-5.3 coding agent on a real project By Flavio Updated Tagged: AI I tried ZCode with GLM-5.3 on a real project. Here's how the coding agent works, what it costs, its permission modes, automations, API, and limits.
model
GLM-5.3-Flash
huggingface.co/zai-org/GLM-5.3-Flash ↗
826875 downloads2200 likesimage-text-to-texttransformers
from the model card
GLM-5.3-Flash 👋 Join our WeChat or Discord community. 📖 Check out the GLM-5.3-Flash blog and GLM-5 Technical report. 📍 Use GLM-5.3-Flash API services on Z.ai API Platform. Introduction We introduce GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series. With 320B total parameters and just 18B active parameters, it outperforms GLM-5.2 across benchmarks and real-world workloads at one-tenth the price, while approaching Claude Opus 4.8 on coding and agentic benchmarks. GLM-5.3-Flash starts from a newly trained base model, with its architecture and training recipe redesigned around capability and efficiency. For the first time in the GLM series, we introduce a hybrid architecture combining sparse and linear attention, sharply reducing long-context serving costs while preserving precise long-context capabilities. The model also adopts Manifold-Constrained Hyper-Connections (mHC) to further improve scaling efficiency. Together with our latest 30T-token multimodal pre-training corpus, these changes enable GLM-5.3-Flash to deliver more intelligence with less compute. Serve GLM-5.3-Flash Locally GLM-5.3-Flash supports deployment with the following frameworks. Feel free to try them out: SGLang — see cookbook vLLM — see recipes TokenSpeed — see here Transformers — see transformers docs KTransformers — see tutorial Unsloth — see guide Note GLM-5.3-Flash supports controlling…
discussions
- Date-fns Migration 3 2026-11-06 – 2026-11-12
- Inference Server Update 2 2026-11-04 – 2026-11-09
- Agent Newsroom Launch 3 2026-11-04 – 2026-11-06
- GLM 5.3 Release 6 2026-10-26 – 2026-11-02
- GLM-5.3 Update 10 2026-10-21 – 2026-10-29
- Claude Sonnet Launch 6 2026-10-11 – 2026-10-17
- GLM 5.3 Release 29 2026-09-09 – 2026-10-09
- GLM 5.3 4 2026-09-06 – 2026-09-11
recent items
ZCode review: Z.ai's GLM-5.3 coding agent on a real project (flaviocopes.com via hn) Our date-fns migration passed every test and invoices still showed up a day early for US customers (www.reddit.com via reddit) Last Tuesday I finally got Moment.js out of our billing app. It was in about 180 files and nobody had wanted to touch it for three years.
Serving GLM-5.3-Flash on one Trainium2 box with NKI layer megakernels (runanywhere.ai via hn) One AWS trn2.48xlarge serves GLM-5.3-Flash at up to 28.9 agent requests a minute on SemiAnalysis's AgentX benchmark, 3.57 million tokens a minute, and takes a million-token prompt to its first token in about three minutes. NKI layer megake…
↯ Glm↯ Date-fns Migration↯ Inference Server Update↯ Agent Newsroom Launch↯ GLM 5.3↯ GLM 5.3glm
Show HN: A live-built agent newsroom for tracking midterms today (midterms2026.noriagentic.com via hn) We set up a bunch of agents to act as a live newsroom for the midterms. There are 6 regional desks tracking parts of the US (mostly sonnet and sol), 2 political partisans (glm 5.3 so they wont get blocked by guardrails), a newsroom head, a…
↯ Glm↯ Inference Server Update↯ Agent Newsroom Launch↯ GLM 5.3↯ GLM 5.3glmsonnet
Built an agent with Claude Code that cleans the negativity out of my Reddit feed, (www.reddit.com via reddit) https://preview.redd.it/36eobjxf1uyh1.png?width=1440&format=png&auto=webp&s=b22b63b1708b06c22f41bb6c5458c1c1b50ff700 Hey everyone, I built Reddit As Intended with Claude Code. It reads my Reddit home feed, hides violence, tragedy, venting…
↯ Glm↯ Agent Newsroom Launch↯ GLM 5.3 Releaseglmopenaiclaude-code
Hermes and REA and GLM 5.3 = Reverse engineering TikTok app for less than $20 (www.scraping.club via hn) A few weeks ago, a TikTok dataset published on Hugging Face caught my eye: it contained metadata of 5.6B videos from the platform. That’s a huge dataset, and the data collection immediately appeared to me to be the most interesting part.
Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses) (blog.cloudflare.com via hn) Cross-request KV cache exposure in the GLM-5.3 Workers AI model, and our remediation with SGLang On September 8, 2026, we were alerted by a security researcher, Gadi Evron from Knostic, to a tenant isolation issue affecting the GLM-5.3 Wor…
↯ Glm↯ Security↯ GLM 5.3 Release↯ GLM-5.3 Update↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3glmsecurity
Why is Kimi 3.1 still not in cursor? (www.reddit.com via reddit) I mean I suspect I know the answer, but before unsubscribing (I spent more than $20k last month in cursor) I wanted to make sure I wasn't missing something The same happened with GLM 5.3 Coding company seems always to have trouble making t…
↯ Glm↯ GLM 5.3 Release↯ GLM-5.3 Update↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3glmcursor
How we test LLM serving changes for coding agents, two hours of GPUs per change (www.upbound.io via hn) Dennis Ramdass Principal AI Engineer AI This series is about drinking our own champagne: we run Modelplane for our own engineers before we ask anyone else to. About 10 of our engineers pilot GLM-5.3, served with Modelplane on 2 nodes of Ne…
↯ Glm↯ GLM 5.3 Release↯ GLM-5.3 Update↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3↯ GLM 5.3glm
GLM 5.3 Flash Is Ace (terfi.fyi via hn) could not extract summary
Claude had a $0.02 AI review its work. The AI found a bug Claude missed. (www.reddit.comhttps) I'm building autonomous bots for the MMO Final Fantasy XI with Claude doing most of the development. I've also had Claude delegate tasks to other models, mainly code reviews, testing, and catching issues before implementation.
Livestream GLM-5.3 playing Minecraft (twitch.sailresearch.com via hn) Agent broadcast This agent runs GLM-5.3 Flash and lives inside a Sailbox, live-streaming over UDP as it runs. A Sailbox is a sandbox that can live forever and is 80–90% more cost-effective than other providers.
GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior (www.lesswrong.com via hn) Authors: Axel Højmark*, Ezra Newman* (*Equal contribution) While reviewing transcripts from standard behavioral evaluations of GLM-5.3, we noticed that the model occasionally reasons, unprompted, about how its actions will shape future mod…
GLM-5.3-Flash TensorFold Pipeline and Converter (github.com via hn) could not extract summary
GLM-5.3 raises researcher fears the Chinese model can fake AI safety tests (www.scmp.com via hn) could not extract summary
Free GLM 5.3 Flash, DSV 4.1 and MiMo 2.6 flash for a month on PGS AI (www.reddit.com via reddit) Some of the absolute best coding models have been hitting, week after week after week. It's been a crazy few months for opensource AI.
GLM-5.3-Flash-REAP144-Coder (github.com via hn) GLM-5.3-Flash-REAP144-Coder A 321B MoE coding model cut in half for home PCs. We kept 144 of the 288 routed experts in every MoE layer of GLM-5.3-Flash, chosen on real agentic coding sessions, then quantized it with an importance matrix bu…
Show HN: EU Podcasts – daily charts from 30 languages (podcasts.eu.org via hn) Yet another project I did with Ox-alpha (glm-5.3-flash) when it was unlimited. I didn't even know there were SO many podcasts and in SO many languages.
Lovable Style Builder via CF Inference GLM-5.3, Kimi K2.7 Code and GLM-5.3 Flash (mainbrella.com via hn) Lovable style Builder via CF Inference GLM-5.3, Kimi K2.7 Code, and GLM-5.3 Flash We’ve built an app builder with the workflow familiar from Lovable: describe what you want, inspect a working preview, and keep talking to the agent to chang…
Write Like It's 1866: LLMs Relearn Telegraphese (fiveminutesforward.com via hn) Everything below is reproducible — harness, frozen ledgers, one-click notebook: github.com/Travis42/telegraph-test. Numbers from a 50-passage, ~1,300-question benchmark: Cross-family matrix (readers = foreign models answering from GLM-5.3-…
Everyone keeps talking about the new open source models. You can try them free on Phoenix Grove API (www.reddit.com via reddit) GLM 5.3 Flash , DSV 4.1 and MiMo 2.6 flash have been making waves, and in some cases even outperforming closed models. If you've been reading the hype and wondering or curious about them, we're giving away a free month of both on Phoenix G…
Show HN: Abliterated LLM provider for cyber tasks (refuseless.com via hn) Current lineup includes GLM 5.3 and its Flash variant. MiMo V2.6 and Qwen Flash Next will be added in the near future.
Anthropic claims popular Chinese AI model has Mythos-class hacking abilities (www.tomshardware.com via hn) Anthropic claims popular Chinese AI model has Mythos-class hacking abilities — frontier red teaming report details weak safeguards on open-weight AI Zhipu AI's GLM-5.3 goes under Anthropic's microscope. Anthropic has released a new report,…
One month coding with GLM 5.3 Flash (wagtail.org via hn) One month on GLM 5.3 Flash 2B tokens later, 🤖 task failed successfully Setting a challenge to spend the whole of September on only one efficient open model felt like a great idea at the time. Turns out not so much in practice.
Ask HN: My agent went haywire, does this output mean anything? (news.ycombinator.com) I have a relatively fresh Hermes agent that I just provisioned (the text below is from its 3-4 chat session ever) running on a headless Debian server. It's using Z.ai's GLM 5.3 Flash model, and I asked it to print details about its local e…
GLM 5.3 flash on AMD GPUs 670 tok/s (twitter.com via hn) we spent september rewriting the kernels behind GLM 5.3 Flash. major release is live on RunInfra today 670 tok/s on Vercel AI Gateway $0.11 per 1M input, $0.45 per 1M output, $0.03 per 1M cached.
CAISI's Assessment of Z.ai's GLM-5.3 Cyber Capabilities (www.nist.gov via hn) The PRC-based company Z.ai (formerly known as Zhipu AI) released a new AI model, GLM-5.3, on August 14, 2026. Z.ai publicly released the model’s weights two weeks later.
Quoting Anthropic Frontier Red Team (simonwillison.net) 29th September 2026 We evaluate several models on 100 tasks from the [internal Binary Exploitation benchmark] (selected at random), and find that GLM-5.3 develops full control flow hijacks in 4% of the trials; Claude Mythos Preview did so…
↯ Glm↯ Security↯ Anthropic Mythos↯ GLM 5.3 Release↯ AI Model Releasered-teamglmmythos+2
Baseten x OpenAI: Use open models like GLM 5.3 and Kimi K3 natively in Codex (twitter.com via hn) Philip Kiely on X: "Enterprise teams can now use open models like GLM-5.3 Flash and Kimi K3 natively in Codex and count spend against their OpenAI commit." @andrewdsouza take a look: Baseten is bringing open-model inference into Codex for…
GLM-5.3 and the spread of advanced cyber capabilities (www.anthropic.com via hn) Subscribe to the Frontier Red Team newsletter Get updates on our latest red-teaming research and findings. Andrew Fasano, Marius Fleischer Cole McFaul, Robert Xiao, Tripp Gallagher Five months ago, we announced Claude Mythos Preview, the f…
↯ Glm↯ Security↯ Anthropic Mythos↯ GLM 5.3 Release↯ AI Model Releasered-teamglmmythos+1