I Ran a Week of Real Open Source Work on Ox Alpha, the Internet's Mystery Free Coding Model

Description and use

The author tested the free anonymous Ox Alpha model over a week on real open‑source engineering tasks. It excelled at bulk rebases, CI fixes, and drafting security reports, but was slower and less up‑to‑date. It is a cost‑effective option for low‑secrecy chores, while high‑risk work should still use named models.

Problem

How to evaluate and use a free, anonymous large‑language model for real open‑source engineering tasks without compromising security or quality.

Outcome

Ox Alpha proved highly effective for high‑volume, low‑secrecy tasks such as rebasing, CI fixes, and draft generation, offering a cost‑free alternative to paid models. It struggled with speed, recent knowledge, and lacked built‑in test generation, but remained a viable fallback for routine engineering chores.

Workflow

• 1. Configure OpenCode to use the preview model via a single line in config.json.
• 2. Run the existing agent pipeline for daily open‑source tasks.
• 3. Monitor outputs for correctness and security claims.
• 4. Review diffs and security reports manually before merging.
• 5. Switch back to a named model for any sensitive or high‑risk work.

Evidence

The model worked through them sequentially, resolving each conflict set and pushing updated branches, and I reviewed diffs afterward instead of hand merging.

Source

I fed our entire backend FastAPI repo with 142 files into OpenCode, and it accurately diagnosed a race condition between our Redis lock and Celery workers without dropping variable names.

Source

Speed is the first tax. Community measurements put it around 40 tokens per second... 12 minutes 36 seconds per prompt.

Source