Tech • AI • Robotics • Game
Fable 5.2 or Opus? When to use each model in Claude Code
7/10TL;DR
Anthropic has positioned Claude Fable 5.2 as its top generally available model for technical work, while recommending Opus 5.5 as the default starting point for most complex coding tasks and Sonnet or Haiku for narrower or cheaper workloads.
KEY POINTS
Fable introduced as the top technical model
Claude Fable 5.2 is being presented as the company’s most capable generally available model for technical fields. It is intended for projects where judgment, accuracy, and end-to-end execution matter more than token cost, especially when problems are large, loosely defined, and span many decisions across a codebase.
Opus 5.5 remains the default choice
For most day-to-day engineering and knowledge work, Opus 5.5 on medium effort is the recommended starting point. It is positioned as the “daily driver” for complex coding, including feature work that touches multiple layers such as data models, APIs, and user interfaces.
Effort level is the first tuning step
When results are not strong enough, the first adjustment recommended is increasing the effort level rather than switching models. That is described as a smaller change that usually preserves the prompt cache, making it a practical way to improve output without fully changing the setup.
Fable is aimed at open-ended migrations
Fable is framed as the better fit for ambitious work such as a full architectural migration of an online shop. In that example, the task includes many milestones and undocumented decisions, with hard constraints that checkout cannot go down and orders cannot be lost, and the model is expected to plan, execute, and test as it proceeds.
Higher quality comes at a higher price
The tradeoff for Fable is cost. It is priced at roughly twice the per-token cost of Opus, making it better suited to work where output quality is considered paramount and where fewer review cycles or less human intervention may offset the higher usage price.
Sonnet targets well-scoped engineering work
For tasks with clear boundaries and a simple definition of done, Sonnet is the recommended step down. It is described as faster than Opus and available at about half the per-token price, making it suitable for bug fixes and other focused implementation work where speed and cost matter.
Haiku serves high-volume, low-cost workloads
Haiku is positioned as the fastest and cheapest model in the family. It is intended for repetitive, high-throughput jobs such as summarization, classification, and routing, including production features like sorting incoming customer messages many times a day through the Claude API.
Multi-model workflows are encouraged
The guidance also supports splitting one project across several models. In a large migration, Fable can act as the lead model while delegating narrower tasks such as package upgrades and test fixes to Sonnet, balancing quality on the core problem with lower cost on routine subproblems.
Subagents expand delegation inside Claude Code
In Claude Code, those delegated units can be handled by subagents, defined as separate Claude instances with their own context. Each subagent works on one part of a job and reports back, allowing a higher-end model to supervise broader strategy while lower-cost models handle scoped execution.
CONCLUSION
The model lineup is being segmented by task shape rather than a single best choice for every job: Opus 5.5 for most complex daily work, Fable 5.2 for open-ended high-stakes projects, Sonnet for focused engineering tasks, and Haiku for scale-sensitive automation. The strategy centers on matching capability, speed, and cost to the level of ambiguity and risk in the work.
Ask a question