Model Selection

Core Principles

1. Match model to complexity, not size — A 50-file refactor that follows a clear pattern is a Sonnet task (high throughput, simple logic). A 3-file architecture decision with trade-offs is an Opus task (deep reasoning). File count and complexity are orthogonal.

2. Sonnet is the workhorse — 80% of .NET development tasks are routine: implement a feature following an established pattern, write tests, fix a known bug, run scaffolding. Sonnet 4.6 handles all of these at higher speed and lower cost.

3. Opus is the architect — Use Opus 4.6 for tasks that require weighing trade-offs, reasoning about system design, debugging subtle issues, or making decisions with incomplete information. Opus excels when the answer isn't obvious.

4. Context window is a budget, not a dumping ground — Sonnet 4.6's large context window enables working with big codebases but doesn't mean you should load everything. Apply context-discipline principles regardless of model. A focused Sonnet session outperforms a bloated one.

5. Haiku for fire-and-forget subagents — When a subagent does a simple lookup, runs a script, or fetches information, Haiku 4.5 is fast and cheap. Reserve heavier models for subagents that need to reason.

Patterns

Task Complexity Assessment

Classify each task to select the right model: