OpenCode v2.2.0 matches Claude Code on RTX 4070 Ti Super

The author tested a fully local AI agent stack running a Qwen mixture-of-experts model on a 16GB GPU against Anthropic's Claude Code. The local setup successfully built a functioning GPU logger with zero help, matching Claude's functional output while costing a fraction of a cent in electricity compared to a $20 subscription. The author argues that for users without flagship model needs, the performance gap between local and cloud agents has narrowed significantly. Open article →