The author shares first impressions of OpenAI's GPT-5.6, comparing it to previous models like GPT-5.5, Opus 4.8, and Fable 5. While an improvement, GPT-5.6 offers incremental gains. A key consideration is its different model sizes: Sol, Terra, and Luna, with Sol being the most advanced. The model also features adjustable reasoning levels, impacting response quality and speed. The author notes that high reasoning settings can rapidly deplete usage limits, especially on subscription plans, and recommends balancing reasoning levels with usage constraints. For instance, using extra high reasoning for planning and medium reasoning for implementation.

Effective use cases include code reviews, where GPT-5.6 is considered capable of replacing most human reviews. For actual code implementation, the author suggests using another model like Fable for planning and then switching to Opus 4.8 for execution. GPT-5.6 also excels at browser navigation and computer use, performing tasks quickly with medium reasoning. The primary technique involves adjusting reasoning levels to manage speed and usage, often starting with higher settings for planning and reducing them for implementation to avoid hitting limits prematurely. Access to relevant tools and data is also crucial for optimal performance.