Mystery Model 'Ox Alpha' Confirmed as Zhipu AI's GLM-5.3-Flash
The mystery model designated as "Ox Alpha" on the LMSYS Chatbot Arena has been officially confirmed to be GLM-5.3-Flash, developed by the Chinese AI company Zhipu AI. This model features multimodal vision capabilities, a 1-million-token context window, and scores approximately 63% on the DeepSWE benchmark. Identifying top-performing mystery models on LMSYS helps the AI community track the progress of frontier models, particularly from Chinese labs competing with Western counterparts. The strong performance of a "Flash" model on complex benchmarks like DeepSWE highlights the rapid advancement of efficient, cost-effective LLMs. GLM-5.3-Flash demonstrates strong coding capabilities with a ~63% score on DeepSWE, a benchmark designed to evaluate coding agents on long-horizon software engineering tasks. Its combination of a massive 1M token context window and multimodal support positions it as a highly competitive lightweight model.
## BACKGROUND
LMSYS Chatbot Arena is a popular crowdsourced benchmarking platform where users compare anonymous LLMs side-by-side to calculate Elo ratings. DeepSWE is a specialized software engineering benchmark created by Datacurve to measure the performance of coding agents on original, long-horizon tasks from active open-source repositories.