Ox Alpha: The Stealth Model Is Live on Cencori

Ox Alpha is live on Cencori.

A frontier-class model with an anonymous operator appeared on OpenRouter this week and immediately became the most probed model on the internet. It's now available through Cencori's AI gateway under its upstream ID, stealth/ox-alpha — completely free while the stealth preview lasts.

Nobody has confirmed who built it. Community fingerprinting (tokenizer behavior, response styling, multimodal handling) points toward Z.ai's GLM 5.x family, with Xiaomi's MiMo team as the other leading theory — but that's speculation, and we won't pretend otherwise.

Early testers are reporting genuinely frontier-tier results in coding and long-horizon agentic work — the kind of output usually reserved for flagship models. Combined with a $0 price tag and a 1M-token context window, it's worth trying today, with eyes open about what "stealth" means for production use.

Reasoning cannot be disabled on this model — the default effort is max. If you need faster responses at lower cost-per-latency, set the effort explicitly to low.

SDK

import { Cencori } from 'cencori';

const cencori = new Cencori({ apiKey: process.env.CENCORI_API_KEY });

const response = await cencori.ai.chat({
  model: 'stealth/ox-alpha',
  messages: [
    { role: 'user', content: 'Design a rate limiter for a multi-tenant API.' }
  ]
});

REST API

curl -X POST https://cencori.com/api/ai/chat \
  -H "CENCORI_API_KEY: $CENCORI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "stealth/ox-alpha",
    "messages": [
      {"role": "user", "content": "Write a Python script to analyze server logs."}
    ]
  }'

• stealth/ox-alpha added to our OpenRouter catalog routing
• Billed to Cencori customers at zero during the preview
• Full support for streaming, tool calling, structured outputs, and reasoning-effort control
• Cencori's security logging, data rules, and observability apply automatically

Read this section before sending anything sensitive:

• The operator is anonymous. OpenRouter knows who they are; you don't. Prompts and completions are retained by the provider but stated not to be used for training. This is not a zero-data-retention model — keep confidential workloads on providers with contractual guarantees.
• The preview can end without notice. When it does, the model ID stops resolving and requests will fail upstream. Treat Ox Alpha as an experiment, not a dependency — route anything critical to a stable fallback model.
• The free pricing dies with the preview. If the model survives into general availability, expect a real price and possibly a new model ID.
• Use the canonical ID stealth/ox-alpha exactly as written — the stealth/ prefix is part of the ID.

• Cencori Models Catalog