Dmitri Lerko conducted an experiment developing an analytical tool during a ten-hour flight using a local AI model on his MacBook Pro M5 Max. The model, Qwen/Qwen3.6-35B-A3B run via LM Studio and OpenCode, performed comparably to cloud models for well-defined tasks but struggled with context exceeding 100,000 tokens. Lerko noted that the cost-effectiveness of local AI depends on the number of attempts and waiting time, contrasting it with the potentially higher per-request cost but faster completion of cloud-based frontier models like Claude Opus. He emphasized that a functional local AI development setup requires not just the model but also pre-loaded Docker images, development environments, and command-line tools. AI
IMPACT Highlights trade-offs in local vs. cloud AI development, influencing choices for developers regarding cost, speed, and offline capabilities.
RANK_REASON Article describes an experiment and analysis of local AI development, not a new release or product launch.
Read on Mastodon — fosstodon.org →
- Boeing 787
- Claude Opus
- Codex
- Dmitri Lerko
- Hacker News
- LM Studio
- loveholidays
- MacBook Pro M5 Max
- OpenCode
- Qwen/Qwen3.6-35B-A3B
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
