Ox alpha lands on openrouter as first stealth model in years, aimed at long-term engineering work

Ox Alpha went live on 20 August as a “stealth” model developed and operated by a third-party provider that chose to remain anonymous during the pre-release phase. OpenRouter functions only as a request router and is explicitly not the developer, owner, or supplier of the model, a distinction the platform stresses. Prompts and completions stay with the provider and are not used for training; all other usage is governed by the stealth-model terms.

The model focuses on code reasoning, continuous agentic work, and production-scale loads. It is intended for long-term software-engineering tasks, complex reasoning, and workflows that blend text with visual context. The context window is one million tokens, and access is free at this stage, with no per-token pricing. The technical sheet indicates a model built from the ground up for long runs rather than one-off queries.

According to platform data, throughput stands at 24 tokens per second (P50, best among providers) and latency at 5.81 seconds (P50, best provider). Reported availability is 99.99 % and 99.15 % respectively, with an automatic fail-over routing mechanism when a top-tier provider fails, provided request filters allow it. Per-provider availability figures are accessible via the Endpoints API.

The combination of a one-million-token context window, reasonable throughput, and zero cost makes Ox Alpha a natural candidate for experiments in workflows that require long context, large-scale refactoring, massive code-base analysis, or tasks that merge technical documentation with diagrams. The conditions stipulate that data remain with the anonymous provider; anyone sensitive to data location or retention policies should consider this before inserting the model into a production pipeline. OpenRouter is soliciting community feedback to guide future development.