#swebench

The article analyzes the evolution of artificial intelligence in software engineering, pointing to...

OpenHands 1.0 resolves 68% of SWE-bench Verified tasks with Qwen3-Coder-480B and adds production-grade Docker sandboxing for self-hosted coding agents.

OpenHands 1.0 résout 68 % des tâches SWE-bench Verified avec Qwen3-Coder-480B et ajoute un sandboxing Docker de production pour agents de codage auto-héber

Deep Dive: NVIDIA Nooa Benchmark Results on SWE‑Bench Verified Explained Quick...

Contaminazione dei dataset, reward hacking e infrastrutture “bucabili”: quando il punteggio non...

High SWE-bench scores like Claude Code's 80.8% don't guarantee real productivity; FutureX's 80% first-pass test rate is the metric that keeps vibe coding sessions flowing.

A benchmark-driven look at SWE-bench and Terminal-Bench to understand whether terminal agents really beat IDE agents on autonomous coding and operational tasks.

A new study finds that 29% of the coding-agent patches that pass SWE-bench Verified keep code the human developer removed, usually by wrapping it in a guard or fallback, and that adding checks for the deletion drops resolution rates from 63.2% to 41.

LLMs don’t fail at hard problems. They fail at the (medium) ones – the ones that require 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴,...

Based on the latest SWE-bench and terminal-bench results, FutureX outperforms Claude Code on specific task types while costing significantly less.

A systematic audit of SWE-bench Verified, the benchmark used everywhere to rank AI coding ability, found that 68 of its 500 tasks link an issue to a pull request that fixes something else, adds unrelated work, or only partly addresses the report.

𝗦𝗲𝗿𝗶𝗲𝘀: 𝗧𝗵𝗲 𝟭.𝟵𝟲% 𝗚𝗮𝗽 - Medium Issues A model cannot learn medium-tier reasoning from one prompt,...