SweepAI has released Sweep Next-Edit 1.5B, a 1.5-billion-parameter model designed to predict the next code edit before it’s made. The model is optimized for local execution, running on a laptop in under 500 milliseconds when using speculative decoding. Despite its relatively small size, it outperforms models more than four times larger on next-edit prediction benchmarks. The model is available in GGUF format with Q8_0 quantization, making it efficient and accessible for local use. It is based on the Qwen2.5-Coder foundation and supports a context length of up to 8,192 tokens, enabling it to understand complex codebases and recent changes effectively. To use the model, users can download the model file and the run_model.py script, then install the required dependencies using pip: uv pip install llama-cpp-python huggingface_hub. After installation, running python run_model.py will launch the model. The model operates using a specific prompt structure that includes file context, recent code changes (diffs), and the current state of the code. This allows it to accurately anticipate the next logical edit. The complete example and implementation details are included in the run_model.py file. For more information, including technical insights and performance benchmarks, users can refer to the accompanying blog post. Additionally, a JetBrains plugin is available to integrate Sweep Next-Edit directly into the IDE, enhancing developer productivity. The model is released under the Apache 2.0 license, making it freely available for both personal and commercial use.