A recent benchmark test on a MacBook Pro with an M4 Pro chip revealed significant performance differences among local coding AI models. The Qwen3.5:35b-a3b-coding-nvfp4 model achieved an impressive 64.10 tokens per second, making it approximately 6.7 times faster than the Qwen3.6:27b model, which managed only 9.56 tokens per second. This speed difference is crucial for practical AI coding agent workflows, where multiple generations and tool calls are common, impacting overall productivity. AI
IMPACT Faster local models can significantly improve the productivity and responsiveness of AI coding agents.
RANK_REASON Benchmark results of local AI models. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Medium — AI coding tag →
- M4 Pro
- GLM-4.7-flash:latest
- Laravel
- Ollama
- Qwen3.5:35b-a3b-coding-nvfp4
- Qwen3.6:27b
- Qwen3-coder-fast:30b
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
