Current code agents fail beyond single-repo fixes; BeyondSWE reveals 45% success on real-world tasks.
Current benchmarks for code agents primarily assess narrow, repository-specific fixes, overlooking critical real-world challenges such as cross-repository reasoning, domain-specialized problem solving, dependency-driven migration, and full-repository generation. To address this gap, we introduce BeyondSWE, a comprehensive benchmark that broadens existing evaluations along two axes - resolution scope and knowledge scope - using 500 real-world instances across four distinct settings. Experimental results reveal a significant capability gap: even frontier models plateau below 45% success, and no single model performs consistently across task types. To systematically investigate the role of external knowledge, we develop SearchSWE, a framework that integrates deep search with coding abilities. Our experiments show that search augmentation yields inconsistent gains and can in some cases degrade performance, highlighting the difficulty of emulating developer-like workflows that interleave search and reasoning during coding tasks. This work offers both a realistic, challenging evaluation benchmark and a flexible framework to advance research toward more capable code agents.
Current code agents achieve only 45% success on real-world tasks beyond single-repository bug fixing, despite high SWE-bench scores.
BeyondSWE benchmark evaluates agents across four dimensions: cross-repo fixes, domain-specific problems, dependency migration, and full system generation.
Search augmentation provides inconsistent performance gains, revealing a critical disconnect between LLM search and coding capabilities.
Existing SWE-bench benchmarks focus narrowly on isolated function-level fixes within single repositories, missing real-world complexity.
Current code agents lack effective integration of search capabilities with coding proficiency for developer-like workflows.
Meituan LongCat Team, Anchun Gui +160
We introduce LongCat-Flash-Thinking-2601, a 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model with superior agentic reasoning capability. LongCat-Flash-Thinking-2601 achieves ...
Tianxin Wei, Ting-Wei Li +27
Reasoning is a fundamental cognitive process underlying inference, problem-solving, and decision-making. While large language models (LLMs) demonstrate strong reasoning capabilities in closed-world se...
Ailin Huang, Ang Li +213
We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: ...
Veuns-Team, Changlong Gao +25
GUI agents have emerged as a powerful paradigm for automating interactions in digital environments, yet achieving both broad generality and consistently strong task performance remains challenging.In ...
Ji Zeng, Dayuan Fu +15
Recently, the frontier of Large Language Model (LLM) capabilities has shifted from single-turn code generation to agentic software engineering-a paradigm where models autonomously navigate, edit, and ...
