GLM 5.3 is Zhipu AI’s newest GLM-5-series flagship, released on 14 August 2026 and positioned as the strongest open-weights coding model available. It is a mixture-of-experts model of roughly 744 billion parameters with about 40 billion active per token.

How is GLM 5.3 different from GLM 5.2?

GLM 5.3 uses the same base model as GLM 5.2, so all improvement comes from post-training rather than a larger or retrained model. Zhipu reports roughly 50% better performance on its internal code benchmark, with gains concentrated in complex coding and long-horizon tasks.

Can I use GLM 5.3 with OpenClaw?

Yes. OpenClaw runs GLM 5.3 through OpenRouter, a direct Z.AI API key, or the GLM Coding Plan endpoint. On OpenClaw Launch, add your key via BYOK and enter GLM 5.3 as a custom model, confirming the current model slug with your provider first.

Are GLM 5.3 weights available?

GLM 5.3 launched through the GLM Coding Plan and API first. Zhipu stated that open weights would follow roughly two weeks after launch, once safety evaluation and hardening were complete.

← Home

Guide

OpenClaw + GLM 5.3: Zhipu AI's New Coding Flagship

Zhipu AI released GLM 5.3 on 14 August 2026, claiming the strongest open-weights coding model currently available. It is an unusual release — the same base model as GLM 5.2, with every gain coming from post-training. Here is what actually changed, and how to run it on an OpenClaw agent.

What GLM 5.3 Is

GLM 5.3 is the newest release in Zhipu AI's (Z.AI) GLM-5 series, succeeding GLM 5.2. It is a mixture-of-experts model of roughly 744 billion total parameters with about 40 billion active per token — the same architecture and the same base weights as GLM 5.2.

That last detail is the story. Zhipu did not train a bigger model; it re-tuned the one it had. All the reported improvement comes from post-training, which is why the release landed quickly after 5.2 and why the gains are concentrated in specific capabilities rather than spread across the board.

What Changed from GLM 5.2

Zhipu reports roughly 50% better performance than GLM 5.2 on its internal Z.ai code benchmark, with the improvement focused on complex coding and long-horizon tasks — the multi-step work where an agent has to hold a plan together across many tool calls rather than answer a single question well.

The result that drew the most attention was security. GLM 5.3 scored 84.5% on CyberGym, a benchmark measuring whether a model can identify and validate security flaws from source code, which Zhipu positions above Anthropic's Mythos at 83.8% and OpenAI's GPT-5.6 Sol at 83.6%. Those are close margins on a single benchmark reported by the vendor, so read them as “competitive at the frontier” rather than as a decisive win.

Worth noting for anyone planning around open weights: GLM 5.3 launched through the GLM Coding Plan and API first, with Zhipu stating that weights would follow roughly two weeks after launch, once safety evaluation and hardening were complete.

GLM 5.3 vs GLM 5.2 vs Kimi K3

Model	Best for	Notes
GLM 5.3	Complex coding, long-horizon agent tasks, security review	Same base as 5.2; post-training gains; leading CyberGym score
GLM 5.2	High-volume Chinese-language chat, cost-sensitive tool calling	Still available and still cheap; fine when coding is not the workload
Kimi K3	Long-document work, general reasoning	Scores higher on general intelligence indexes than the GLM line

Pick GLM 5.3 when your agent writes or reviews code, or runs long multi-step tasks. Stay on GLM 5.2 if your workload is conversational and cost per token is the binding constraint — the 5.3 improvements are concentrated where 5.2 was already adequate for chat. Reach for Kimi K3 for long-document reasoning.

How to Deploy GLM 5.3 with OpenClaw Launch

GLM 5.3 is new, so you run it with your own key (BYOK) rather than through the curated one-click models. It still deploys in about 30 seconds:

Go to openclawlaunch.com and open the configurator.
Add your OpenRouter key in the BYOK settings, or bring a direct Z.AI API key. If you subscribe to the GLM Coding Plan, its endpoint works here too.
In the model dropdown, choose Custom model and enter the GLM 5.3 model ID. Confirm the exact slug on openrouter.ai/models or in Zhipu AI's docs — provider model IDs shift between releases and a stale slug fails as a model-not-found error.
Pick your chat platform (Telegram, Discord, WhatsApp, WeChat, or the browser gateway) and click Deploy.

Tip: Switch between GLM 5.3, GLM 5.2, Kimi K3 and DeepSeek any time from the dashboard or with the /model chat command — no redeploy needed. A common pattern is a cheap model for everyday chat and GLM 5.3 for the coding and review work.

Self-Hosted Configuration

Running OpenClaw on your own server? Configure the provider in your openclaw.json and point the default agent at GLM 5.3:

{
  "models": {
    "providers": {
      "openrouter": {
        "apiKey": "sk-or-..."
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "openrouter/z-ai/glm-5.3"
      }
    }
  }
}

Verify the current slug before deploying rather than copying it from any guide, including this one. If you hold a direct Z.AI key instead, configure Z.AI as an OpenAI-compatible provider with its own base URL — see the custom endpoint guide.

When the Security Benchmark Actually Matters

A strong CyberGym score means the model is good at reading code and spotting flaws in it. If your agent reviews diffs, audits dependencies, or triages reports, that is directly useful. If it answers questions in a group chat, it is a headline that will not change your day. Worth pairing with our agent security guide either way — a model that can find vulnerabilities is not the same as an agent deployment that is safe from them.

What's Next?

OpenClaw + GLM 5.2 — the previous flagship, and still the cheaper choice for chat
OpenClaw + GLM — the general GLM guide covering the wider GLM-4 and GLM-5 lineup
Hermes + GLM — run GLM models on Hermes instead of OpenClaw
OpenClaw + Kimi K3
OpenRouter setup guide
Compare all models