Nominal 
The Economic News Agency reported on the 24th that an anonymous AI model named '0x Alpha' has quietly launched on the OpenRouter and OpenCode platforms, immediately igniting the global developer community. The model not only topped OpenRouter on its first day, breaking the single-day usage record, but also ended DeepSeek's 56-day consecutive reign at the top of OpenCode. Based on its technical characteristics, the market highly suspects that this model is very likely a variant of Zhipu (02513)'s GLM series.
Platform data shows that 0x Alpha is specifically designed for code writing, long-duration agent tasks, and real-world production environments. Its standout feature is an ultra-long context window of up to 1,048,600 tokens, with a maximum output length of 131,000 tokens, and full support for multimodal reasoning across text, images, and video. In comparison, the context processing capabilities of current mainstream large language models generally range between 8,000 and 128,000 tokens.
Amid the surge in popularity, OpenRouter's page shows that the cumulative usage of the top five entry points has exceeded 40 trillion tokens; OpenCode has also announced it has prepared a daily service capacity of 100 trillion tokens, enabling global developers to conduct large-scale real-world stress testing.
Regarding the model's identity, the developer community, through technical investigations such as tokenizer fingerprint identification and API response pattern matching, highly suspects it to be a variant of Zhipu AI (Z.ai)'s GLM-5.3, although Zhipu has not yet made any official confirmation. Market analysis suggests that if its domestic origin is ultimately confirmed, it would signify that Chinese-developed large models have not only approached the performance of top overseas models in benchmark testing (Benchmark), but also demonstrated scalable, real-world capabilities in handling trillions of token calls and supporting global agent workloads. (nw)
Platform data shows that 0x Alpha is specifically designed for code writing, long-duration agent tasks, and real-world production environments. Its standout feature is an ultra-long context window of up to 1,048,600 tokens, with a maximum output length of 131,000 tokens, and full support for multimodal reasoning across text, images, and video. In comparison, the context processing capabilities of current mainstream large language models generally range between 8,000 and 128,000 tokens.
Amid the surge in popularity, OpenRouter's page shows that the cumulative usage of the top five entry points has exceeded 40 trillion tokens; OpenCode has also announced it has prepared a daily service capacity of 100 trillion tokens, enabling global developers to conduct large-scale real-world stress testing.
Regarding the model's identity, the developer community, through technical investigations such as tokenizer fingerprint identification and API response pattern matching, highly suspects it to be a variant of Zhipu AI (Z.ai)'s GLM-5.3, although Zhipu has not yet made any official confirmation. Market analysis suggests that if its domestic origin is ultimately confirmed, it would signify that Chinese-developed large models have not only approached the performance of top overseas models in benchmark testing (Benchmark), but also demonstrated scalable, real-world capabilities in handling trillions of token calls and supporting global agent workloads. (nw)