Matse says Ox Alpha found many bugs and security holes in a year's worth of code and fixed multiple problems in about three hours, but provides no sample or repair evidence; it is a candidate audit workflow, not a performance conclusion.
Code scope: The author's “one year of work”; no repository or language is given.
Comparison: The author had previously used MiniMax; no parallel run on the same codebase is published.
Time: About three hours.
Outcome: The author says the model found “endless Bugs and Security holes” and solved or improved many problems.
The post gives no bug count, severity distribution, CWE, reproduction steps, diff, test pass rate, false-positive rate, or human review. “Endless” and “world's leading” are the author's wording and cannot be converted into a count or ranking.
Fix the codebase by module and commit, and establish a baseline with existing tests and static scans.
Ask Ox Alpha to report the file/line, CWE, impact, reproduction steps, and minimal fix for every finding.
Put fixes on a separate branch and run tests, SAST/dependency scans, and a human security review.
Repeat with MiniMax or another fixed-version baseline, recording true positives, false positives, missed issues, side effects, and duration.
This is a useful positive experience to turn into a controlled security-audit test. Without the audit artifacts, it cannot establish Ox Alpha's security capability or superiority over MiniMax.
Ox Alpha
