An anonymous provider has been distributing free access to a frontier-class AI coding model through OpenRouter Inc. since August 20, and no company has come forward to take credit for building it.
Access and Specifications
The model, listed as Ox Alpha, carries no cost for input or output tokens. OpenCode, an open-source terminal agent, introduced the model on the same date and promoted a capacity of 100 trillion tokens per day with rate limits described as nearly unlimited. OpenRouter told users the free access on its route runs through August 24. OpenCode’s launch note promised a week of access from August 20, suggesting the free window extends to approximately August 27.
The model’s context window reaches 1,048,576 tokens with output capped at 131,072. It accepts text, images, and video. Audio requests return rejected.
OpenRouter’s listing positions the model as a reasoning system designed for long-horizon software engineering and sustained agentic work.
Benchmark Performance
Early benchmark scores drew attention. Developer Ben Davis, posting as @davis7, ran Ox Alpha through 10 tasks on the DeepSWE software engineering benchmark and scored it above 80 percent. Claude Fable 5 managed 65 percent on the same tasks. GPT-5.6-sol achieved 52 percent. Davis flagged his own small sample size, noting that results from a subset that narrow carry substantial variance.
That caution proved warranted. After running the full DeepSWE set, Davis found Ox Alpha performed roughly on par with GPT-5.6-sol rather than significantly ahead of it. No public leaderboard has published a result for the model.
Attribution Investigations
The effort to identify the builder has attracted more scrutiny than the benchmark testing. Developer unclecode, who created the Crawl4AI open-source crawler, built a browser tool called modelprint that fingerprints anonymous API endpoints. The tool sends nine infrastructure probes to an endpoint and matches the responses against known models.
Ox Alpha aligned with GLM-5.3 on six of the nine probes, including all four normalized tokenizer counts. No other lab’s best candidate matched more than two of those four.
Unclecode has been clear about what that proves. “Matching fingerprints prove shared infrastructure, not identity,” he wrote in the project documentation. He notes that a lab can serve two different models on the same stack.
Potential Sources
Z.ai Co. has a track record with anonymous previews. The company previously released GLM-5 on OpenRouter under the name Pony Alpha ahead of its public debut. GLM-5.3 shipped August 14, six days before Ox Alpha appeared.
Other testers have proposed different candidates. Xiaomi Corp.‘s MiMo team has come up in discussions and has previewed unbranded models before. One analysis of the tokenizer behavior points toward cl100k_base, an encoding developed by OpenAI Group PBC that sits awkwardly with a Chinese attribution.
Data Retention and Compliance Concerns
The terms of service have drawn scrutiny alongside the benchmarks. OpenRouter’s model page states that prompts and completions are retained by the provider and not used for training. The broader terms covering anonymous previews on the platform extend to training, evaluation, and improvement. OpenCode’s route advertises zero retention from a provider it does not name.
That gap carries weight if the Z.ai connection holds. The U.S. Commerce Department added Zhipu AI to its Entity List on January 16, 2025, under a rule stating that listed companies advance China’s military modernization through developing and integrating advanced AI research. Zhipu AI is Z.ai’s former name.
Enterprise code is flowing to a provider that has no public identity. Users sending that code cannot verify where it ultimately lands.
Despite the questions, usage has not slowed. Coding tools including Claude Code have pushed billions of tokens through the model since Thursday, according to Bloomberg. OpenRouter says it is not the developer, owner, or operator of Ox Alpha and only routes requests to it.