Here is small guide to help you decide how to choose the model for the tasks at hand:
First of all, understand there are 2 types of models:
- Thinking (they have a brain symbol adjacent to the name)
- Non-Thinking
Furthermore, these models have effort capabilities:
- High Effort (requires significant computational resources, time, and specialized expertise, generally resulting in higher accuracy, better reasoning, or more specialized capabilities)
- Max Effort (leads to maximum tokens and computational resources to generate the highest quality output, prioritizing accuracy over speed or cost
Opus 4.6 (Non-Thinking, High Effort ) Use this model for: Everyday knowledge work, general coding, document/spreadsheet/presentation tasks, research, financial analysis, and conversational use. This most workloads. The model uses “adaptive thinking” here. It decides on its own when deeper reasoning would help, without you needing to toggle it on. Great balance of quality, cost, and speed.
Opus 4.6 (Non-Thinking, Max Effort) Use this model for: Deep agentic search, hard-to-find information retrieval, and tasks requiring exhaustive exploration without structured reasoning traces. Choose this when the task is about breadth, persistence, and tool use over many steps and not about chain-of-thought reasoning on a single hard problem.
Opus 4.6 (Thinking, High Effort) Use this model for: Complex multi-step coding, debugging, navigating large codebases, code review, and general hard reasoning tasks where you want the model to think carefully but don’t need absolute peak performance. The model thinks more deeply and revisits its reasoning before answering.
Opus 4.6 (Thinking, Max Effort) Use this model for: The absolute hardest problems i.e. frontier-level reasoning challenges, complex multidisciplinary questions, advanced math/science, and tasks where accuracy matters more than cost or speed. This is the most expensive and slowest configuration, but it pushes to the ceiling of what the model can do.
Quick rule of thumb: If the task is about searching, browsing, or executing many tool calls, use non-thinking + max effort. If the task is about reasoning through a hard problem, use thinking + max effort. For everything else, the default (high effort, adaptive thinking) is the right choice.
