OpenChat

Open-source model matching ChatGPT performance using C-RLFT

Code & DevelopmentFree

OpenChat is an open-source language model that achieves ChatGPT-level performance using Conditioned Reinforcement Learning Fine-Tuning (C-RLFT)—a novel training approach using mixed-quality data without expensive human feedback. OpenChat 3.5 7B surpassed ChatGPT 3.5 on multiple benchmarks at release, demonstrating that efficient training strategies can match larger proprietary models. OpenChat models run efficiently on consumer hardware via Ollama and llama.cpp.

Key Features

• ✓C-RLFT training
• ✓ChatGPT-competitive
• ✓7B parameter size
• ✓Ollama compatible
• ✓Fast inference
• ✓Open weights