Momentic launches Mo, an autonomous QA agent for finding software bugs
Reported by Hacker News and aiweekly2 sources ↓
Momentic introduced Mo, an agent that tests apps by exploring their flows. In Momentic’s QA Bench, Mo matched 269 of 1,396 human-labeled bugs; a Hacker News post describes a login change that skipped new-device verification.
Key points
- Mo matched 269 of 1,396 labeled bugs in Momentic’s QA Bench
- Momentic reported 84% precision and a cost of $2.67 per confirmed bug
- Codex was the next-best configuration, matching 86 bugs in the benchmark
- The benchmark covered four open-source apps and 20 product areas
Sources
Details
AI Weekly reports that Mo was tested against Codex, Claude Code and Devin on Gitea, Metabase, Twenty and Saleor. Momentic’s benchmark assigned 86 bugs to Codex at 86% precision, 59 to Claude Code and 27 to Devin; the reported cost for Codex was $0.80 per bug.
The QA Bench uses Momentic’s own hand-labeled set of 1,396 bugs, rather than an independently assembled benchmark. Momentic says Mo is built on three years of test traces and uses parallel sub-agents to explore app controls and user flows. A Hacker News post describes a consumer fintech app where a login-routing change bypassed new-device verification.
Written from coverage as of
Images
- Hacker News