Models & Labs
Z.ai Unveils Powerful Open-Weight AI Model GLM 5.3
WIRED AIhigh confidence
Why it matters
- →GLM 5.3 offers a cost-effective solution for cybersecurity tasks, potentially democratizing access to advanced AI capabilities.
- →The model's open-weight nature allows for broader experimentation and use, but also poses dual-use risks.
- →This release underscores China's growing influence in AI development, challenging US dominance in the field.
Chinese AI company Z.ai has announced the release of GLM 5.3, a powerful open-weight model designed for coding and cybersecurity tasks. This model is said to perform on par with leading models from Anthropic and OpenAI, offering a cost-effective solution for identifying system vulnerabilities. While the model is currently in limited release to trusted partners, its capabilities raise concerns about potential misuse by cybercriminals. The release highlights China's growing influence in the development of open-weight AI models, despite US efforts to limit access to advanced AI training chips.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1OpenAI Launches GPT-5.5 for Cybersecurity
OpenAI · May 7, 2026 · Related
- 2GLM 5.2 Emerges as Leading Open Weights Model
Sam Witteveen · June 17, 2026 · Same story
- 3GLM 5.2 Released Amid AI Ecosystem Shifts
The AI Daily Brief · June 21, 2026 · Related
- 4GLM 5.2 Analysis Highlights Open-Weight Model Impact
The AI Daily Brief · June 22, 2026 · Same story
- 5Zhipu AI's GLM-5.2 rivals Mythos in cybersecurity
The Verge AI · June 28, 2026 · Same story
- 6Z.ai Releases GLM-5.2 Open-Source Model
Matt Wolfe · July 1, 2026 · Same story
- 7OpenAI's AI System Breaches Hugging Face Security
The Verge AI · July 21, 2026 · Related
- 8OpenAI Models Hack Hugging Face in Security Test
WIRED AI · July 25, 2026 · Related
What happened next
- 9GLM-5.3 Outperforms Mythos 5 in Cybersecurity
Lev Selector · August 21, 2026 · Same story
- 10Z.ai Revealed as Creator of Ox Alpha Model
TechCrunch AI · August 26, 2026 · Related
- 11Z AI Reveals Ox Alpha as GLM-5.3-Flash
The Rundown AI · August 27, 2026 · Related
- 12GLM 5.3 Flash: Z.AI's Cost-Efficient Model Variant
Sam Witteveen · August 30, 2026 · Related
- 13Reflection debuts Beam open-weight model
TechCrunch AI · October 5, 2026 · Related
Follow this story
Open the full story →Open-weight AI models near top-tier capabilities
3 developments
- Aug 4 · TechCrunch AI
Open-weight AI models near top-tier capabilities - Aug 18 · WIRED AI
Z.ai Unveils Powerful Open-Weight AI Model GLM 5.3 (This article) - Aug 18 · AI News
Zhipu's GLM-5.3 excels in cybersecurity benchmark↳ GLM-5.3 scored 84.5% on CyberGym but trails Anthropic's Mythos 5 on ExploitBench and ExploitGym.
More from WIRED AI
© WIRED AIAgentsagents
Meta's Muse Agent Builds Detailed Profiles of Your Social Circle
Meta’s new AI assistant, Muse, is quietly building comprehensive dossiers on everyone in your life. Researchers extracted system prompts revealing an automated process that creates individual pages for friends, family, and colleagues, tracking everything from birthdays to relationship dynamics. This goes far beyond simple memory; it attempts to model the nuance of human connections to offer proactive advice on strengthening ties. The approach raises significant privacy concerns, as the agent infers details from your interactions rather than just storing explicit data. It marks a shift toward AI that understands social context with unsettling depth.
WIRED AI
© WIRED AIInvestment · $40-100M
General AIresearch
Trillium Labs launches open AI research nonprofit
Nathan Lambert and Tom Zick are launching Trillium Labs to challenge the closed-door model of frontier AI safety. Backed by Schmidt Sciences and aiming for $40-100M in funding, the nonprofit will publish detailed experiments on recursive self-improvement and reinforcement learning. This moves high-stakes safety research from proprietary labs into the open scientific method, allowing external scrutiny of how models behave under pressure. It signals a growing institutional demand for transparency in AI development.
WIRED AI
© WIRED AIMarket & Regulationother
ChatGPT Mac App Vulnerability Exposes User Data
A critical flaw in the ChatGPT macOS app allowed local malware to bypass security checks and hijack the application. Researchers at Objective-See found that a script interpreter could be tricked into executing untrusted commands, granting attackers access to chat logs and browser sessions. The exploit was trivial, requiring only a dozen lines of code to spoof process lineage. OpenAI patched the issue after public disclosure, highlighting the risks of deep system integration in AI tools. This incident underscores how feature expansion can inadvertently widen the attack surface for end-user applications.
WIRED AI
More in Models & Labs
Models & Labscoding
vLLM v0.31.0: SM100 defaults and fast restarts
vLLM is quietly becoming the definitive runtime for NVIDIA's latest hardware, making NVFP4 compressed KV caches the default for DeepSeek-V4.1-Flash on SM100 GPUs. This isn't just a performance tweak; it fundamentally changes how enterprise inference scales by keeping post-quantized weights resident in GPU memory across engine restarts via the new preload daemon. The release also hardens speculative decoding with Model Runner V2, fixing OOMs that previously plagued wide expert deployments. For builders, this means lower latency and higher throughput on next-gen hardware without manual configuration overhead.
vLLM Releases
Models & Labsmodels
llama.cpp v0.6.0 adds decision model API and Apple Silicon speedups
This release shifts llama.cpp from a pure text engine to a multimodal inference runtime capable of handling 'decision models' like Clef and GLM-5.3-Flash. The new /v1/systemone server endpoint standardizes how these non-autoregressive models are queried, while the extended batch API allows mixed token and embedding inputs for complex architectures. Apple Silicon users get a tangible performance boost with new Metal MMA kernels that accelerate speculative decoding by up to 3x. It’s a significant step toward supporting the next generation of hybrid reasoning models locally.
llama.cpp Releases
© Hugging Face BlogModels & Labsmodels
Falcon-Emirati-7B targets dialect nuance
Most Arabic models treat the language as a monolith, missing the cultural and linguistic depth of specific dialects. Falcon-Emirati-7B closes this gap by fine-tuning on native Emirati text, synthetic data constrained by strict glossaries, and cultural heritage knowledge. It tops the new Alyah benchmark with 84.83%, proving that scale alone doesn't buy dialect competence. This release underscores a critical shift: true multilingual capability requires targeted adaptation, not just larger parameter counts.
Hugging Face Blog