A mysterious new model, codenamed Ox Alpha, launched on OpenRouter and OpenCode and surged to the top of OpenRouter's leaderboard within its first day, setting a new record for single-day model usage.

According to OpenRouter's page, the combined usage from the top five entry points, including Hermes and Claude Code, has already exceeded 4 trillion tokens. Ox Alpha also ended DeepSeek's 56-day consecutive streak at the top of OpenCode, becoming the go-to "model" for global developers over the past two days.

Stripe CEO Patrick Collison praised the model as "very impressive," while the Hermes community has also given positive feedback. An official evaluation from the DeepSWE team suggests a test score of around 63, placing its overall performance close to Claude Opus 4.8 and within the ranks of leading global high-performance models.

Several Google DeepMind researchers have posted hints suggesting this is a new version of Gemini, fueling rapid speculation that Ox Alpha is linked to a new Google model. However, the community has pointed to tokenizer fingerprints and error codes as evidence, steering the attribution toward Chinese AI firm Z.AI's GLM series. As of now, Z.AI has not confirmed the model's ownership.

Notably, OpenCode has announced it has prepared a service capacity of 100T tokens per day for Ox Alpha, enabling global developers to conduct near-unlimited real-world stress testing. If the model is confirmed to be of Chinese origin, it would suggest that domestic large models are not only approaching overseas frontier performance in benchmarks but may also already possess the capability to handle trillion-token calls and support global agent workloads at scale.