Zhipu (Z.ai) releases GLM-5.3, top open-weights coding/agent model
Confirmed
The takeaway
Z.ai (Zhipu AI) released GLM-5.3 on 2026-08-14 via its coding service, a post-training upgrade of the GLM-5 base (753B parameters) that it calls the most capable open-weights coding model, with weights published on Hugging Face about two weeks later after an extended risk review.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 13 October 2026
Your AI and this story
- GPT-6 Astra106 days after its cutoff
- Claude Opus 5.545 days after its cutoff
- Gemini 4 Hydrogen136 days after its cutoff
- Grok 4.775 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 45 days before it.
Key facts
- 753B parameters; same base model as GLM-5.2, gains from post-training only (Hugging Face model card)
- Terminal-Bench 3.0: 28.3 (up from 4.6 for GLM-5.2); DeepSWE 66.9 (from 46.2); SWE-Marathon 42.5 (from 19.4)
- HLE with tools 62.5; CyberGym 84.5; Agents’ Last Exam 28.5
- Claimed +50% over GLM-5.2 on Z.ai Code Bench
- Weights on Hugging Face (zai-org/GLM-5.3) around 2026-08-28 under a custom GLM-5.3 license
- Series context: GLM-5 (Feb 2026), GLM-5.1 (Apr), GLM-5.2 (June 13, MIT license)
- GLM-5.3-Prime (OpenRouter z-ai/glm-5.3-prime, listed Sept 23, 2026, slug dated 2026-09-21): high-speed variant with ‘1.5–2x the output throughput through inference acceleration’, text-only, 1M context, 131,072 max output, reasoning always on; $2.80/$8.80 per 1M input/output tokens on OpenRouter
What happened
GLM-5.3 first shipped on 2026-08-14 through Z.ai’s coding plan/API, with Zhipu committing to open weights ~two weeks later following what it called its most extensive risk review (the model scores highly on offensive-cyber benchmarks such as CyberGym and ExploitBench). The model card reports large jumps on long-horizon agentic coding benchmarks. Fortune reported that when Hugging Face was breached by OpenAI’s evaluation agents in July, it used a Z.ai open model for defensive analysis.
Why it matters
Zhipu, which listed in Hong Kong in January, shows Chinese open models competing at the top on agentic coding. The staged release (API first, weights after risk review) is an emerging norm for dual-use-capable open models.
Update, Oct 13, 2026: GLM-5.3-Highspeed in limited test
Pandaily (Oct 13, citing QbitAI) reports that a model named GLM-5.3-Highspeed appeared in Zhipu’s coding tool ZCode (version 3.15.1) as a gray release to randomly selected subscribers. One user’s screenshot showed about 490 tokens per second; Zhipu has not announced the model or published a speed figure. For comparison, GLM-5.3-FlashX peaked near 200 tokens/s and Artificial Analysis measured GLM-5.3-Flash on the Z.ai API at about 60 tokens/s. The open-source coding agent Pi already lists the model with a 1M-token context, 131,072 max output and low/high/max reasoning effort. Reddit users speculate it is “Omen Alpha”, an anonymous model seen on OpenCode in September; that is unconfirmed. The model is not on Z.ai’s pricing page or release notes (checked Oct 13).
Sources
5 sources from 5 sites. Numbers match the chips in the text.
5 sources: 2 primary, 3 press
Primary
- Hugging Face: zai-org/GLM-5.3huggingface.co, official
- OpenRouter: GLM 5.3 vs GLM 5.3 Primeopenrouter.ai, docs
Press
- Pandaily: Zhipu’s GLM-5.3-Highspeed Appears in ZCode Gray Release, With Users Reporting About 490 Tokens/spandaily.com, press
- MLQ: Zhipu releases GLM-5.3 through its coding servicemlq.ai, press
- Emergent: GLM-5.3 officially launchedemergent.sh, press
Changes
- Added the GLM-5.3-Highspeed gray release in ZCode (Pandaily/QbitAI; unconfirmed by Zhipu)
- Added GLM-5.3-Prime high-speed variant (OpenRouter listing, found while checking for GLM-5.4)
- Filed