GLM 5.3 Release Date

TL;DR — GLM 5.3 is already here. Z.ai launched the GLM-5.3 flagship on August 14, 2026, keeping the GLM-5.2 base model and delivering all improvements through post-training. It is live in the GLM Coding Plan, the public API is coming soon, and open weights are expected around August 28, 2026 after a safety review. Specs: 1M-token context, 128K max output, text-in/text-out, reasoning always enabled.

The short answer: GLM 5.3 shipped on August 14, 2026

If you have been tracking "when does GLM 5.3 come out," the wait is over. Z.ai published the official GLM-5.3 model page and launch announcement on August 14, 2026, naming it the new flagship in the GLM-5 family — a date multiple independent outlets confirm.

That date matters because most pages ranking for "GLM 5.3 release date" were written in July, when the model was still a rumor — "not out yet," "expected soon." That framing is now outdated: the model is live, and the open questions have shifted to when the API and open weights will land, not whether the model exists.

What is GLM 5.3?

GLM 5.3 is Z.ai's (Zhipu AI's) latest flagship LLM, aimed at complex software engineering and long-horizon agent work. The key fact: Z.ai did not train a new base model. Per the official documentation, GLM-5.3 uses the same base model as GLM-5.2, with all gains coming from an expanded post-training phase run longer and across more realistic, long-horizon task environments.

Confirmed specs from the official model page:

• Context window: 1M tokens
• Maximum output: 128K tokens
• Modality: text in, text out (no documented vision or multimodal input)
• Reasoning: always enabled, with three effort levels — low, high, and max

One migration note if you already build on GLM APIs: disabling the thinking layer is no longer supported. Applications that previously set thinking.type: "disabled" must switch to enabled and set reasoning_effort before moving to the glm-5.3 model ID, otherwise requests will fail.

GLM 5.3 release timeline: how we got here

Z.ai has moved through the GLM-5 generation quickly, and the cadence frames what "release date" means here — the announcement, subscription rollout, API, and open weights each land at different times.

The pattern to notice: GLM-5.3's "release date" is not a single moment. The model is usable now through the GLM Coding Plan, which serves it in tools such as Claude Code, Cline, and OpenCode. The public API is "coming soon," and the open weights are expected roughly two weeks after launch — around August 28, 2026 — held back for safety evaluation and hardening, the first GLM-series release explicitly delayed for that reason.

What changed vs. GLM 5.2

Because the base model is unchanged, every improvement is a post-training result. Z.ai reports a roughly 50% gain on its in-house Z.ai Code Bench over GLM-5.2, plus open-source SOTA results on public coding benchmarks. The reported numbers from the official model page:

On CyberGym, a vulnerability-discovery benchmark, Z.ai reports 84.5%, ahead of Anthropic's Mythos 5 (83.8%) and OpenAI's GPT-5.6 Sol (83.6%). Z.ai also says the model is more token-efficient: at High effort it reaches 31.4% on Code Bench at roughly 50K output tokens per task, versus Claude Opus 4.8 at 29.5% using about 120K.

One caveat: these are vendor-reported figures run in Z.ai's own harness — a starting point for evaluation, not a guarantee. Independent verification only becomes possible once the open weights are released.

The cybersecurity capability Z.ai did not plan for

The unusual part of this launch is security. Z.ai introduced vulnerability-discovery data into post-training expecting incremental improvement. Instead, as training scaled, the model began reasoning across multiple stages of exploitation and forming coherent attack-chain plans — a capability Z.ai says grew faster than expected.

This is not just a benchmark story. Working with security teams including NSFOCUS, CyberKunlun, and DARKNAVY, plus researchers from Tsinghua and Nankai universities, Z.ai reports its models have identified 2,436 vulnerabilities across 269 real-world projects since GLM-5.2, of which 1,097 are rated medium-to-high severity. Findings are tracked in the public Z.ai Security Disclosure Ledger at cvd.z.ai, with the oldest bugs dating back roughly 40 years.

For most developers the practical takeaway is simpler: if you use GLM 5.3 for security review, treat it as an assistant layer that experts still validate — open weights and licensing are not finalized yet.

How to use GLM 5.3 today

You do not need to wait for the API to try the model:

1. Chat with it. Start a free GLM 5.3 chat on glm5.app, describe your problem and constraints, then iterate with your own code and acceptance criteria.
2. Subscribe to the GLM Coding Plan. This is the only official way to run GLM-5.3 inside coding tools right now — it works with Claude Code, Cline, and OpenCode. Existing subscribers were upgraded to the new model.
3. Integrate when the API goes live. Z.ai has not published the API date or pricing yet. For OpenAI-compatible access, expect the model ID glm-5.3; glm5.app's endpoint (https://glm5.app/api/v1) exposes a Chat Completions interface.

Bottom line

The GLM 5.3 release date is August 14, 2026 — and unlike earlier GLM-5 patches, this is a post-training jump on the same base, not a new architecture. Try it now through the GLM Coding Plan or a free chat, watch for the API announcement, and mark late August for the open weights if you plan to self-host.

FAQ

What is the GLM 5.3 release date? GLM 5.3 launched on August 14, 2026, when Z.ai published the official model page, specs, and benchmark results.

Is GLM 5.3 available now? Yes, in limited form. It is fully rolled out in the GLM Coding Plan and available for chat on glm5.app. The public model API is listed as "coming soon."

When will GLM 5.3 open weights be released? Z.ai says the weights will follow roughly two weeks after launch, placing the expected date around August 28, 2026, pending safety evaluation. This is a plan, not a confirmed date.

What is the difference between GLM 5.3 and GLM 5.2? The base model is the same. GLM-5.3 adds a much larger post-training phase, reporting about a 50% coding gain on Z.ai Code Bench and big jumps on Terminal-Bench 3.0, DeepSWE v1.1, and cyber benchmarks.

Does GLM 5.3 support images or video? No. The official documentation lists GLM-5.3 as text-in/text-out.

How much does GLM 5.3 cost? Z.ai has not published separate API pricing. The GLM Coding Plan is a subscription, and API pricing and limits will only be known when Z.ai announces them.

Sources

• Z.ai Developer Docs — GLM-5.3 Overview — Official specs, availability, benchmarks, reasoning parameters, and Security Disclosure Ledger details.
• Z.ai Blog — GLM-5.3: Frontier Coding with Emergent Cyber Capabilities — Official launch announcement dated August 14, 2026.
• GLM Coding Plan — Official subscription page where GLM-5.3 is currently served.
• TechTimes — GLM-5.3 launch coverage — Independent reporting on the August 14 launch and the planned ~August 28 open-weights release.
• Z.ai Blog — GLM-5.2: Built for Long-Horizon Tasks — Source for the June 16, 2026 GLM-5.2 release and its open-weight precedent.

Note on sourcing: benchmark figures are vendor-reported by Z.ai in its own harness and may change as independent verification and release announcements land. Availability, pricing, and open-weights timing are subject to change — confirm on the official Z.ai pages before relying on them.

Related Articles

• Colibri AI with GLM 5.2: Enhancing Real-Time AI Conversations - GLM 5
• DeepSeek V4 Flash vs GLM 5.2: Cheap Speed or Depth? - GLM 5
• GLM 5.2 Agentic Workflows: Function Calling, Tool Use, and Multi-Step Tasks - GLM 5