A new AI model called Ox Alpha just appeared on OpenRouter with no company name attached, and the internet is already trying to solve the mystery.
The model launched with free access, a one-million-token context window, and multimodal input. It is built for coding, sustained agentic work, and production workloads. Within hours, developers were running tests and comparing it against the best systems from OpenAI, Anthropic, and Google.
Early benchmark results caught everyone off guard. Ox Alpha scored 80% on a DeepSWE subset, a coding benchmark that measures how well AI can solve real software engineering problems. More complete testing put it at 63%, placing it near Fable 5 while using far fewer tokens per task. That kind of efficiency is rare among frontier models.
The bigger puzzle is who built it. Digital detectives examined the model’s answers and its naming convention, which follows a Chinese zodiac theme. Those clues point to China’s Zhipu AI, potentially a version like glm-5.3 flash or glm-6. Another possibility is Microsoft’s MAI family, although the last four anonymous drops on OpenRouter in six months all came from Chinese labs.
Ox Alpha is drawing massive usage because the provider is offering near-unlimited free access for the week, with capacity for 100 trillion tokens a day. That kind of scale lets developers stress-test the model in ways that usually cost hundreds of dollars.
Why it matters
We have never seen a smaller model compete so directly with frontier systems. If Ox Alpha can run locally on consumer hardware, near-frontier coding will no longer need the cloud. That would change how developers build software, how startups compete, and how organisations think about data sovereignty.
We still need the full reveal to confirm its origins and capabilities. If it is small enough to run locally, it will break the internet in the best way possible.
The race for accessible, powerful AI just got more interesting.
Related Reading
- OpenAI Claims a $1M Millennium Prize With a Secret Model. The Credit Fight Is Only Beginning
- DeepMind Chief Proposes U.S. AI Watchdog Before Frontier Models Hit the Market
- Frontier AI agents took unauthorised actions on the live internet during UK safety tests
The views expressed on this site are my own and do not represent those of any current or former employer. Articles are based on publicly available information and are provided for general educational purposes.
