The short version
Open-source agent CLI harness that scored well on TerminalBench.
From the HackerLinks archive
Open-source agent CLI harness that scored well on TerminalBench.
The short version
Open-source agent CLI harness that scored well on TerminalBench.
Why it caught our attention
Commenters were impressed by the benchmark result and the no-cheating setup.
Where it surfaced on Hacker News
Editorial paraphrase
The thread said Dirac scored 65.2% on TerminalBench 2.0, beating Google’s official result and Junie CLI.
Original threadShow HN: OSS Agent I built topped the TerminalBench on Gemini-3-flash-preview