Quick Facts
- Ox Alpha launched anonymously on OpenRouter on Aug. 20, 2026, offering a 1,048,576-token context window at zero cost for one week.
- Forensic analysis points to Zhipu AI’s GLM-5.3 as the likely source, matching on 30 of 30 tokenizer probes, though attribution remains unconfirmed.
- Zhipu AI, which operates as Z.ai, was added to the U.S. Commerce Department Entity List in January 2025 over military AI concerns.
An AI coding model called Ox Alpha appeared on OpenRouter on Aug. 20, 2026, with no company name, no technical paper, and no announcement. The model was described on the platform as a reasoning model built for coding, agentic work, and production workloads. The anonymous provider claimed capacity of 100 trillion tokens per day.
OpenCode, an open-source coding agent, listed Ox Alpha as free for one week, through approximately Aug. 27. The model supports text, image, and video inputs, with a context window of 1,048,576 tokens and a maximum completion of 131,072 tokens.
Developer Ben Davis ran Ox Alpha through DeepSWE, a coding-agent benchmark, and initially reported a score of 80%, ahead of other models. A fuller run later placed it roughly level with GPT-5.6-sol mid. No public leaderboard has posted an independent result.
Stripe CEO Patrick Collison, whose company is acquiring OpenRouter, called Ox Alpha “very impressive” on X. Usage accelerated quickly. Bloomberg reported Sunday that coding tools including Claude Code had pushed billions of tokens through the model since its release.
The Identity Investigation
Community investigators ran a coordinated forensic effort to identify the model’s origin. A developer known as unclecode built a browser tool called modelprint that sends nine infrastructure probes to an API endpoint and matches responses against known models. Ox Alpha matched GLM-5.3, a model developed by Chinese company Zhipu AI, on six of nine probes, including all four normalized tokenizer counts.
Researcher @aitrackerbot ran 25 prompts through Ox Alpha and recorded exact token counts. A follow-up by researcher Chetaslua expanded this to 30 probe strings covering 14 writing systems, emoji, code, and SQL. Every prompt matched GLM-5.3’s tokenizer. Each request showed a constant offset of 75 tokens, consistent with a system prompt or routing wrapper added invisibly to every query.
On Aug. 22, serving-layer forensics identified a Java stack trace and error code 1214, both consistent with Zhipu AI’s Z.ai infrastructure. OpenRouter has not confirmed the attribution. Unclecode cautioned: “Matching fingerprints prove shared infrastructure, not identity.”
Other analysts pointed to Xiaomi’s MiMo team, which previously deployed stealth models on OpenRouter under pseudonyms before official announcements. No statement from any company has confirmed the model’s origin.
The Privacy and Regulatory Gap
OpenCode’s documentation states Ox Alpha follows a zero-retention policy and does not use customer data for training. That policy is not an independent audit, a public data-processing agreement, or verified by any named legal entity. Users cannot confirm where their code goes because no provider has identified itself.
The gap carries regulatory weight if the Zhipu attribution holds. The U.S. Commerce Department added Zhipu AI to its Entity List on Jan. 16, 2025, citing the company’s role in advancing China’s military modernization through AI research. China’s National Intelligence Law requires Chinese organizations to cooperate with state intelligence work.
Enterprise developers are sending production code, internal logic, and proprietary systems through a model with no named owner. Security teams should treat Ox Alpha as an unverified third-party API and avoid routing sensitive source code, credentials, customer data, or internal documents through the service until the provider identifies itself and a verified data-processing agreement exists.
Read more: Nobody knows who built AI coding model Ox Alpha or where the code goes