Fable 5.2 or Opus? When to use each model in Claude Code
Start coding work on Opus; use Fable for ambitious, open-ended work and cheaper models for well-scoped or high-volume tasks.
- Try raising Opus effort before switching models; this is a smaller change and usually keeps the prompt cache.
- Use Fable when a large, open-ended task needs strong judgment and accuracy; state the goal and non-negotiables.
- Use Sonnet for clear, bounded tasks and Haiku for high-volume work such as sorting or classifying messages.
The model-selection guidance and examples are clear from the transcript; no detailed walkthrough is needed. · Best for: Developers choosing Claude models for coding tasks and workflows.
Choose a model based on task scope, required quality, speed, and cost. Use Opus as the default, then move up or down based on the work.
Start with Opus for complex coding0:00
- Opus 5.5 on medium effort is the recommended starting point for complex coding and knowledge work.
- For a feature spanning data, API, and UI, begin with the default model and effort.
- If results fall short, raise effort before changing models; this usually preserves the prompt cache.
Use Fable for ambitious work1:00
- Choose Fable for broad, loosely defined projects that require judgment and accuracy.
- It costs twice as much per token as Opus, so reserve it for work where quality is paramount.
- Provide the overall goal and critical constraints; the example constraints are keeping checkout available and not losing orders.
Use Sonnet or Haiku for bounded work1:00
- Sonnet suits well-scoped tasks with clear completion criteria; it is faster than Opus and half its per-token price.
- Haiku is the fastest and cheapest option, suited to high-volume, cost-sensitive tasks such as summarization, classification, and routing.
- The example uses Haiku through the Claude API to sort incoming customer messages.
Split large jobs across models2:07
- Use Fable to lead a complex migration, then delegate bounded subtasks such as package upgrades and test fixes to Sonnet subagents.
- A Claude Code subagent has its own context, handles one part of the work, and reports back.
Measure on your own work3:16
- Use the guide mentioned below the video to measure how each model performs on your tasks and projects.
Step by step
- Start complex coding work with Opus at medium effort.
- If the result is insufficient, increase the effort level before switching models.
- For broad, open-ended work, move to Fable and specify the goal and non-negotiable constraints.
- For clearly bounded work, use Sonnet; for high-volume, cost-sensitive tasks, use Haiku.
- For complex projects with separable subtasks, have the main session delegate bounded work to subagents using an appropriate model.
- Measure model performance on your own projects using the mentioned guide.
Tools & resources mentioned
- Claude Code — main coding session and subagent delegation.
- Claude API — runs the example Haiku message-sorting feature.
- Guide linked below the video — steps for measuring model performance on your own tasks and projects; no URL is provided in the transcript.
Notes are AI-generated from the video's transcript and may miss nuance.