NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·3h ago
System Design
Martin Fowler·29d ago·5 min read
Viability of local models for coding
TL;DR
Local models for coding have been developed to enhance software development efficiency. These models leverage specific coding patterns and context to generate relevant code snippets.
✦ Why It Matters
Engineers can implement local coding models today to reduce development time and improve code quality in their projects.
Key Takeaways
- Test local models on machines with sufficient RAM for better performance.
- Use Qwen3.6 35B MoE for a balance of speed and capability.
- Be prepared for configuration challenges when integrating local models.
- Evaluate the specific task requirements to choose the right model.
Full Summary
Recent advancements in local coding models have made them more viable for developers, particularly for agentic coding, which involves more than just auto-completion. The author tested various models on Apple M3 Max and M5 Pro machines, focusing on factors like RAM, processing power, and tool calling capabilities.
The Qwen3.6 35B MoE model emerged as the most effective, balancing performance and resource usage. Despite improvements, issues with tool calling and the need for extensive configuration persist, making it less user-friendly for those seeking a straightforward experience.
The findings suggest that while local models are promising, they are not yet ready for plug-and-play use.