(6/7) The agentic approach mirrors developer workflows: broad exploration → context building → converging to root cause. Combines efficiency of specialized retrievers with depth of agentic reasoning.

📄 arxiv.org/abs/2512.20482

(5/7) Results show significant improvements: → SWE-PolyBench: 58.28% Acc@10 → SWE-Bench-Multilingual: 64.10% Acc@10 → Multi-turn reasoning: 3-6 point boost over single-pass → Effective for low-overlap cases

📄 arxiv.org/abs/2512.20482

(4/7) SWERANKAGENT uses iterative, multi-turn reasoning for complex bugs: → Progressive searches across turns → Memory buffer of relevant functions → Query refinement from intermediate results → Evidence aggregation before final ranking

(3/7) SWERANKMULTI extends code ranking to 10 languages using SWELOCMULTI dataset (155K instances, 4K+ repos): → Bi-encoder retriever → LLM reranker → JavaScript, Java, TypeScript, Ruby, Rust, Go, PHP, C, C++, Python

📄 arxiv.org/abs/2512.20482

(1/7) SWERANK+: Multilingual, Multi-Turn Code Ranking for Software Issue Localization

First multilingual framework for issue localization across 10 programming languages.

📄 arxiv.org/abs/2512.20482

#FutureOfAI #EnterpriseAI

SFR-DeepResearch agents beat OpenAI on Humanity's Last Exam

Our SFR-DR-20B achieved 28.7% on Humanity's Last Exam—outperforming OpenAI's o3-powered Deep Research on PhD-level questions through innovative RL training.

📖 bit.ly/4nDvmut

#EGI #DeepResearch #AIAgents #EnterpriseAI

Join a team where research meets real-world impact. Salesforce AI Research advances AI techniques for NLP, computer vision, and multimodal systems—publishing at top conferences while building products used by millions.

Explore opportunities: bit.ly/41rEH0S

Welcome to Salesforce AI Research!

We bridge cutting-edge AI research with enterprise solutions — from Large Action Models to multimodal AI and sustainable architectures.

Explore: www.salesforceairesearch.com/publications

#AIResearch #EnterpriseAI