A user with a 32GB system asks if open-weight models can match Opus 4.8's 1M context and coding performance on local hardware. They note current bottlenecks are context length and privacy concerns, and question whether high-end models like GLM 5.2 or Qwen3.7 are feasible within a $3.5K budget, emphasizing that running 70-80B models offers marginal real-world gains over 27B models with 256K context.