Show HN: OSS Agent I built topped TerminalBench on Gemini-3-flash-preview | Hacker News

An open-source terminal agent reportedly scored 65.2% on TerminalBench, topping Google’s reported 47.8% and Junie CLI’s 64.3%.

The author said the result was achieved using the published code as-is, with no hidden modifications or cheating.

The claim underscores how benchmark setup can materially change agent performance rankings.

It also renews concerns about benchmark integrity, comparability, and leaderboard trust.