research

ReviewBenchLite: A Benchmark for Evaluating Automated Code Review Capabilities of Language Models

We introduce ReviewBenchLite, a benchmark for systematically evaluating the code review capabilities of language models and autonomous agents. Unlike existing benchmarks that focus on code generation or bug fixing given explicit problem descriptions, ReviewBenchLite tests the ability to proactively identify issues in production codebases without prior knowledge of what problems exist.

Omnigrep: Agentic Code Search via Multi-Turn Chain-of-Thought Reasoning

Code search represents a critical bottleneck in AI-assisted software engineering pipelines. When autonomous coding agents are tasked with repository-level modifications such as debugging, feature implementation, or refactoring, they must first locate relevant code spans before any downstream reasoning can occur.