Dev Tools · 1h ago
Local AI Coding Agent Mitii Hits 78% Success on 515 Tasks
Mitii, a local AI coding assistant using Qwen3-Coder 30B, achieved a 78% pass rate across 515 tasks including bug fixes and security tests. The agent's Ask mode scored 87% on hard tasks, while Plan mode reached 87% on easy tasks. The benchmark proves local LLMs can be production-ready for code generation.
Meridian48 take
The results are promising for local AI coding, but the 78% overall rate and variability across modes suggest developers should still exercise caution before fully trusting autonomous agents with production code.
Read the full reporting
Benchmarking Mitii AI Agent: 78% Success Rate on 500+ Tasks Using a Local Qwen3-Coder (30B) →
DEV Community
ai-coding-agentlocal-llm-benchmark