The mysterious Ox Alpha is finally unmasked as Z.ai's GLM-5.3-Flash
The ghost model that dominated OpenCode and OpenRouter under the pseudonym Ox Alpha has finally revealed its true identity. Z.ai just dropped GLM-5.3-Flash, and it is supposedly ready to crush every leaderboard in sight.
Z.ai has officially unveiled GLM-5.3-Flash, the first natively multimodal member of the GLM-5 family. Boasting a massive 320 billion parameters, the model is designed to activate only 18 billion during active generation, keeping things surprisingly lean for such a heavy hitter. It handles up to 1 million tokens of context, positioning itself as a direct contender for high-volume development tasks.
Before this big reveal, the model roamed the wild as Ox Alpha, an anonymous entity that climbed to the top of OpenCode and OpenRouter rankings in just six days. According to reported data, it processed an staggering 42 trillion tokens during that period. The model effectively doubled the throughput of its closest rival, DeepSeek V4 Flash, in weekly rankings.
The weights are already public on Hugging Face under an MIT License, meaning developers can start playing with the engine immediately. Z.ai claims superior performance over its predecessor, GLM-5.2, on various coding and agentic benchmarks. However, the company is relying on its own internal benchmarks like Code Bench, where it landed just shy of Claude Opus 4.8. While it reportedly handled 63.4% on DeepSWE, the lack of independent third-party testing makes these numbers feel like a well-optimized highlight reel.
Perhaps the most curious detail is the infrastructure. Z.ai claims the entire Ox Alpha traffic surge was handled by a massive cluster of tens of thousands of Chinese-made accelerators. The specific hardware remains a company secret, leaving the industry to guess whether they found a magic chip or just a really good pile of electricity.
This is the ultimate performance art of the AI world: release an anonymous beast, dominate the leaderboards to build hype, and then pull back the curtain to reveal a corporate product. Whether this is a genuine technological leap or just a masterclass in marketing manipulation remains the burning question, especially since nobody has yet seen these results replicated outside of the company's own walled garden.
Source: Z.ai
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.