Artificial intelligence field has once again seen a breakout product making a stunning debut. AI model aggregation platform OpenRouter officially announced on August 20 that it has launched a new anonymous model codenamed Ox Alpha. Although the developer's identity remains mysterious, the model has quickly sparked intense discussions in overseas technical communities due to its outstanding performance in multiple benchmark tests.

According to the information released by the platform, Ox Alpha is defined as a cutting-edge AI model focusing on code writing, long-term agent tasks, and real production environments. In terms of core parameters and technical indicators, it demonstrates an extremely high ceiling: its context window capacity reaches approximately 1048000 tokens, with a maximum output length of 131000 tokens, and it fully supports multimodal input including text, images, and videos. Currently, it offers free access for developers during the preview period. At the same time, the open-source terminal programming agent OpenCode also announced on the same day that it has integrated this model, preparing an impressive daily call capacity of up to 100 trillion tokens.

Within just a few hours of its launch, Ox Alpha attracted a large number of developers for intensive evaluations. Independent researcher Ben Davis conducted strict testing on 10 sub-tasks from the software engineering benchmark DeepSWE. The results showed that Ox Alpha achieved a pass rate of about 80%, outperforming many well-known industry models, including Claude Fable 5 with a 65% pass rate, GLM-5.3 and Grok 4.6 with 62%, and GPT-5.6 Sol with 52%. However, industry insiders also pointed out that this set of data currently comes from preliminary testing by an individual researcher, and the DeepSWE official leaderboard has not yet officially included this model.