GLM 5.3 Release Date & What's New: Complete Timeline

GLM 5.3 was released on August 14, 2026. The API is live at $1.40/$4.40 per 1M tokens, and open weights landed on Hugging Face on August 25. Full timeline and what changed vs GLM 5.2.

GLM 5.3 Release Date & What's New: Complete Timeline

Quick answer: Z.AI released GLM 5.3 on August 14, 2026. It went to the GLM Coding Plan and ZCode on launch day. OpenRouter listed it on August 18, and the open weights followed on Hugging Face on August 25. Every access route is now open: Z.AI's API costs $1.40 input and $4.40 output per 1M tokens. GLM 5.3 uses the same base model as GLM 5.2, and every gain comes from post-training: +50% on coding, open-weights SOTA on Terminal-Bench 3.0, and an emergent cyber capability.


TL;DR

MilestoneDate
Official releaseAugust 14, 2026
Coding Plan and ZCodeAugust 14, 2026 (all subscribers upgraded)
OpenRouter listing (z-ai/glm-5.3)August 18, 2026
Open weights on Hugging Face (zai-org/GLM-5.3)August 25, 2026
ZCode 1.5× quota boostEnded August 31, 2026
Z.AI APILive, $1.40 input / $4.40 output per 1M tokens

Release Timeline

August 14, 2026: Launch day

Z.AI announced GLM 5.3 in a blog post titled "GLM-5.3: Frontier Coding with Emergent Cyber Capabilities." The headline facts:

  • Same base model as GLM 5.2. Every gain comes from scaled post-training.
  • +50% coding capability on Z.AI's internal code bench.
  • Open-weights SOTA on Terminal-Bench 3.0 (28.3, against 4.6 for GLM 5.2) and Agents' Last Exam.
  • CyberGym 84.5, the best public vulnerability-discovery result at launch, ahead of GPT-5.6 Sol (83.6).
  • 2,436 real-world vulnerabilities found across 269 projects during testing.

GLM Coding Plan subscribers were upgraded automatically, and ZCode had it the same day. On launch day, the API was announced as "coming soon," with its endpoints already published.

August 18, 2026: OpenRouter listing

OpenRouter's catalog dates the z-ai/glm-5.3 listing to August 18, four days after launch. It now runs on dozens of provider endpoints. See GLM 5.3 on OpenRouter for current prices and setup.

August 25, 2026: Open weights on Hugging Face

Z.AI had said the weights would follow about two weeks after launch, once safety evaluation and hardening were done, because GLM 5.3's cyber capability is dual-use. They arrived on August 25, slightly ahead of that schedule:

  • zai-org/GLM-5.3: FP8 weights, about 753B parameters in total.
  • zai-org/GLM-5.3-BF16: the same model in BF16.
  • License: a custom glm-5.3 license rather than a standard open-source one. Read it before commercial use.
  • Serving: the model card points to SGLang and vLLM, with Ascend NPU support through vLLM-Ascend, xLLM, and SGLang.

GLM 5.3 Flash was published on Hugging Face the same day as zai-org/GLM-5.3-Flash.

August 31, 2026: ZCode boost ended

ZCode's limited-time 1.5× quota boost ran until August 31. It no longer applies.

What's New vs GLM 5.2

AreaWhat changed
Coding+50% on Z.AI Code Bench; Terminal-Bench 3.0 from 4.6 to 28.3; DeepSWE from 46.2 to 66.9
AgentsAutomationBench from 26.2 to 48.2; ALE-CLI from 23.8 to 28.5
Cyber (emergent)CyberGym from 77.2 to 84.5 (SOTA); ExploitBench from 24.4 to 54.4
Efficiency34.5% at about 75K tokens, against 23.4% at about 96K for GLM 5.2 (Max effort)
ThinkingAlways on, with reasoning_effort set to low, high, or max; disabled was removed
Context1M-token context window, up to 128K output tokens
InputText only

How to Use GLM 5.3 Today

  1. Z.AI API: use the model ID glm-5.3. It costs $1.40 input, $0.26 cached input, and $4.40 output per 1M tokens. Thinking is always enabled, so remove any thinking.type: "disabled" from GLM 5.2 code. The migration guide covers the details.
  2. OpenRouter: use z-ai/glm-5.3 to reach many providers with one key. Prices vary widely by endpoint, so compare them in our OpenRouter guide.
  3. GLM Coding Plan: subscribe at z.ai/subscribe to use it in ZCode, Claude Code, Cline, OpenCode, and other coding agents. Calls made off-peak, including all weekend, use half the standard points.
  4. Self-host: download zai-org/GLM-5.3 from Hugging Face and serve it with SGLang or vLLM. At around 753B parameters, it needs multi-GPU datacenter hardware.

For a full price comparison, see our GLM 5.3 pricing guide.

FAQ

When was GLM 5.3 released? On August 14, 2026, through the GLM Coding Plan and ZCode. The OpenRouter listing followed on August 18 and the open weights on August 25.

Are GLM 5.3's weights open? Yes. Z.AI published zai-org/GLM-5.3 (FP8) and zai-org/GLM-5.3-BF16 on Hugging Face on August 25, 2026, under a custom glm-5.3 license.

Is GLM 5.3 available through the API? Yes. Use the model ID glm-5.3 on Z.AI's API at $1.40 input and $4.40 output per 1M tokens, or z-ai/glm-5.3 on OpenRouter.

What's the biggest change in GLM 5.3? It keeps GLM 5.2's base model but has much stronger post-training: +50% on coding, Terminal-Bench 3.0 from 4.6 to 28.3, and an emergent cyber capability that tops CyberGym.

Can I still get the ZCode 1.5× boost? No. That offer ended on August 31, 2026.


Sources

Last updated: November 11, 2026

Start Using GLM 5 Today

Try GLM 5 free — reasoning, coding, agents, and image generation in one platform.