Ox-Alpha Revealed as GLM-5.3-Flash

Back to Devoured - August 27, 2026

DEVOURED

Ox-Alpha Revealed as GLM-5.3-Flash

The viral 'ox-alpha' model is revealed as Z.ai's GLM-5.3-Flash, a 320B-parameter MoE model optimized for ultra-low-cost inference.

What: Chinese lab Z.ai identified 'ox-alpha' as their new GLM-5.3-Flash model, which features 320B total parameters and 18B active parameters. It claims parity with Claude Opus 4.8 on coding and agentic tasks while being designed for high-throughput efficiency.

Deep dive

320B total parameters, 18B active parameters (Mixture of Experts architecture).
Claims performance on par with GPT-5.6 Terra and Gemini 3.7 Flash.
Architected specifically for low-cost inference using hybrid attention.
Served entirely on Chinese AI chips.
Weights are available on Hugging Face.

Decoder

MoE (Mixture of Experts): A model architecture where only a subset of the model's parameters ('experts') are activated for any given input, significantly reducing compute cost without losing capacity.

Original article

Z.ai revealed that the anonymously tested ox-alpha was GLM-5.3-Flash, a 320B-parameter MoE model with 18B active parameters that approached Claude Opus 4.8 on coding and agentic benchmarks.

Explore all 59 articles in today's digest