Kodezi Chronos

Stale
GitHub Java NOASSERTION

Description

Kodezi Chronos is a debugging-first language model achieving state-of-the-art performance on SWE-bench, capable of autonomously handling software debugging and code repair tasks.

Key Features

  • Debugging-first language model trained on 42.5M real debugging examples
  • 80.33% success rate on SWE-bench Lite benchmark, leading all systems by 20pp
  • Persistent Debug Memory (PDM) learns from 15M+ debugging sessions across projects
  • 89% root cause accuracy with multi-file patch generation
  • 7-layer architecture optimized specifically for repository-scale code understanding
  • 67.3% autonomous debugging success rate, 4-5x improvement over GPT-4.1

Use Cases

💡 Autonomous software debugging and bug fixing in large codebases
💡 Repository-level code analysis and root cause identification
💡 Multi-file patch generation for complex logic and concurrency bugs
💡 Enterprise-grade code quality assurance and regression prevention
💡 Research benchmarking for debugging capabilities of language models

Quick Start

Kodezi Chronos is proprietary and available through Kodezi OS. Join the waitlist at chronos.so for early enterprise access (Q4 2025 beta, Q1 2026 GA). This repository contains research paper, benchmarks, and evaluation results. Install via: access the model through Kodezi OS platform.

Related Projects