Qwen 3.6 35B A3B shows local coding models are already enough

A Reddit user has sparked a viral discussion by claiming that the Qwen 3.6 35B A3B local LLM is already good enough for serious coding, technical planning, and hardware setup. According to the post, the model only failed when the user hadn't taken the time to set up proper tooling, direction, and context. With a sound plan in place, the model 'hasn't skipped a beat' and handles every task without struggle.

The post explicitly notes that this is not about casual use—it's about real development work. The user argues that any perceived shortcomings are actually workflow or discipline gaps, not model limitations. This raises a provocative question: are we simply demanding mind-reading from LLMs, and are further improvements just enabling laziness?

This sentiment resonates with developers who run models locally to avoid cloud costs, latency, and privacy concerns. The Qwen 3.6 35B A3B is a 35-billion parameter model with a mixture-of-experts architecture (A3B likely referring to a specific sparse activation design). It competes with other local models like Llama 3 and Mistral but is noted for its efficiency and reliability once properly integrated. The discussion highlights a growing belief that current-generation local models may already satisfy most practical coding needs, challenging the rush toward ever-larger models.