Frontier Models
Ox Alpha Stealth Model Emerges on OpenRouter as GLM-5.3 Variant
Anonymous release on August 20, 2026, features 1M context and multimodal inputs for coding agents, with stylometric fingerprints and benchmark results indicating a Zhipu AI GLM-5.x origin amid ongoing Chinese lab stealth tests.
Ox Alpha is a stealth frontier model from provider Stealth added to OpenRouter on August 20, 2026, featuring a 1,048,576-token context window and multimodal capabilities for coding and agent tasks.
The emergence of Ox Alpha on OpenRouter represents the latest in a series of stealth frontier model tests by Chinese AI laboratories.
Background on OpenRouter Stealth Releases
On August 20, 2026, an anonymous frontier model called Ox Alpha appeared on the platform.
It came from provider Stealth and quickly saw real production traffic from various agents and tools.
Previous similar stealth models on the platform resolved to Chinese labs including Zhipu GLM-5, Xiaomi MiMo, Ant Group, and Meituan models.
Core Features of the New Model
Ox Alpha features a 1,048,576-token context window.
The model supports text, image, and video inputs while generating text outputs.
It is offered free during the preview period.
Positioning for Agentic and Coding Workloads
The model is designed for coding, sustained agentic work, and production workloads.
It suits long-horizon software engineering and complex reasoning tasks.
Workflows that combine text with visual context are supported.
Fingerprinting Points to Zhipu GLM-5.x
Independent researcher Ben Davis is 99% certain the model belongs to Zhipu's GLM-5.x series.
Evidence includes video encoder token consumption patterns identical to GLM-5V-Turbo.
Additional indicators are tokenizer alignment with GLM-5.3 and audio input rejection behavior.
Output style such as emoji usage rates also matches.
Ox Alpha is a reasoning model designed for coding, sustained agentic work, and production workloads. It is suited for long-horizon software engineering, complex reasoning, and workflows that combine text with visual context.OpenRouter Platform listing
Evidence from Prior Stealth Models
Previous similar stealth models on the platform resolved to Chinese labs including Zhipu GLM-5, Xiaomi MiMo, Ant Group, and Meituan models.
- Video encoder token consumption patterns identical to GLM-5V-Turbo
- Tokenizer alignment with GLM-5.3
- Audio input rejection behavior
- Emoji usage rates in output style
Market and Stakeholder Implications
The quick appearance of production traffic suggests immediate utility in real-world applications.
Agents like Hermes Agent and tools like Claude Code and ZCode have utilized the model.
| Aspect | Detail |
|---|---|
| Context Window | 1,048,576 tokens |
| Input Modalities | Text, image, video |
| Output | Text |
| Pricing | Free preview |
| Positioning | Long-horizon coding and agents |
Link to GLM-5.3 Development
Z.ai describes GLM-5.3 as having stronger coding capabilities with a 50% improvement over GLM-5.2.
The 1-million-token context window aligns with the described long-horizon task post-training.
What's Next for Frontier Model Testing
This pattern indicates continued anonymous testing by Chinese labs on public platforms.
Further releases may follow similar trajectories with advanced capabilities.
Frequently asked
What evidence links Ox Alpha to Zhipu AI?
Fingerprinting evidence from video encoder patterns, tokenizer alignment, output style, and audio rejection behavior points to a GLM-5.x variant from Zhipu AI, with researcher Ben Davis stating 99% certainty.
When did Ox Alpha launch and what are its key specs?
Ox Alpha launched on August 20, 2026, on OpenRouter with a 1,048,576-token context window, supporting text, image, and video inputs, and is free during the preview period.
Sources
- OpenRouter — Ox Alpha is a reasoning model designed for coding, sustained agentic work, and production workloads with a 1,048,576-token context window, released Aug 20, 2026, free, accepts text, images and video.
- OrcaRouter — On August 20, an anonymous frontier model called Ox Alpha appeared with a 1-million-token context window, text/image/video input, free for the next week, following pattern of previous stealth models claimed by Chinese labs including Zhipu AI's GLM-5.
- Z.ai — GLM-5.3 is the most capable open-weights model for coding, with a 50% improvement over GLM-5.2 and a 1-million-token context window for long-horizon task post-training.
- BigGo Finance — Ox Alpha achieved 80% Pass@1 rate on DeepSWE coding benchmark subset (small sample of ~10 tasks), and Davis stated he is 99% certain the model belongs to Zhipu's GLM-5.x series.