Hacker News: ibragim_bad

13 Models and 4 Agents on SWE Tasks: Go, Java, Python, Rust, TS

ibragim_bad — Fri, 31 Jul 2026 15:28:28 +0000

Article URL: https://swe-rebench.com

Comments URL: https://news.ycombinator.com/item?id=49124336

Points: 41

# Comments: 15

SWE-Rebench (December 2025)

ibragim_bad — Fri, 16 Jan 2026 20:32:04 +0000

Article URL: https://swe-rebench.com/?insight=dec_2025

Comments URL: https://news.ycombinator.com/item?id=46651820

Points: 1

# Comments: 0

New comment by ibragim_bad in "SWE-rebench: Over 21,000 Open Tasks for SWE LLMs"

ibragim_bad — Thu, 29 May 2025 11:59:02 +0000

Hi! We just released SWE-rebench – an extended and improved version of our previous dataset with GitHub issue-solving tasks.

One common limitation in such datasets is that they usually don’t have many tasks, and they come from only a small number of repositories. For example, in the original SWE-bench there are 2,000+ tasks from just 18 repos. This mostly happens because researchers install each project manually and then collect the tasks.

We automated and scaled this process, so we were able to collect 21,000+ tasks from over 3,400 repositories.

SWE-rebench: Over 21,000 Open Tasks for SWE LLMs

ibragim_bad — Thu, 29 May 2025 11:59:02 +0000

Article URL: https://huggingface.co/datasets/nebius/SWE-rebench

Comments URL: https://news.ycombinator.com/item?id=44125129

Points: 5

# Comments: 1