An anonymous lab quietly deployed Ox Alpha on OpenRouter under the stealth/ox-alpha endpoint, pitching a multimodal reasoning engine aimed squarely at complex software engineering, visual pipelines, and autonomous agent loops. The release gives developers a massive 1M-token context window at zero compute cost, with OpenRouter acting as a direct passthrough proxy to a single undisclosed host.
Stealth drops like this have effectively transformed public API aggregators into blind firing ranges. By routing live production queries through third-party gateways, frontier AI labs stress-test unreleased architectures on high-entropy developer traffic without putting brand reputations on the line. According to OpenRouter's telemetry, the deployment clocks a median latency of 4.39 seconds, an inference throughput of 23 tokens per second, and a three-day uptime of 99.46%.
Zero-dollar token economics always come with fine print. Under OpenRouter's Stealth Model Terms, the anonymous vendor retains all submitted prompts and completions, though they explicitly commit not to use them for downstream model training. For enterprise engineering teams, that makes sending proprietary codebases or sensitive business context a clear compliance hazard. Free compute is enticing for long-horizon synthetic benchmarks, but piping closed-source production logic through an unverified black box remains an unforced security gamble.