Transcript
0:00 Claude Fable 5.2 is out. It's our most capable generally available model for technical fields. With several models to choose from, how do you pick the right one for the task in front of you? Most of the time, the answer is Opus 5.5 on medium effort. It's the daily driver for complex coding and knowledge work, and the starting point we recommend. As an example, I am working on a online shop and I would like to add monthly subscriptions.
0:27 That change touches the data model, the API, and the UI. I'll leave the model and effort level on the default and start the task. If the result falls short, the first thing to try is raising the effort level. It's a smaller step than changing the model, and it usually keeps the prompt cache. Some work, however, is more ambitious than a single feature. I'd like to migrate the whole shop to a new architecture. It's a large, loosely defined problem across the whole codebase, with dozens of milestones and decisions that nobody has written down yet.
1:00 For work that is ambitious and open-ended, where judgment and accuracy matter, move to Fable. At twice the price of Opus per token, it's suited to work where quality is paramount. I'll give Fable the goal and two non-negotiables. Checkout can't go down and orders can't be lost. It makes the plan, does the work, and tests as it goes, so the result is much closer to what I consider done when I review it. In the other direction, for well-scoped work where speed and cost matters, consider moving down to a faster, cheaper model.
1:36 Let's say a bug report comes in for the shop with a clear reproduction. I can say what done looks like in one sentence. Sonnet is suited to well-scoped tasks like this one, and it's faster than Opus at half the price per token. For even simpler work, you can move further down to Haiku. It's the fastest and cheapest of the family, and it's well suited to high-volume, cost-sensitive work like summarization, classification, and routing.
2:07 For example, the shop has an inbox for customer messages, and I can add a feature that sorts each one as it comes in. That feature runs on Haiku through the Claude API many times a day. You can also split one job across many models. Let's say the shop has to move to a new major version of its framework across every package. It's a large, complex migration, so I'll use Fable, but not every part of it needs its capabilities.
2:37 Upgrading each package and fixing its tests is a well-scoped task. In Claude Code, the main session can hand parts like these to subagents. A subagent is a separate Claude with its own context. It does one part of the job and reports back. So I've asked Fable to lead the migration and hand those parts to Sonnet. Alright, hopefully that gives you a sense of where each model fits. Most tasks start on Opus, the daily driver for complex coding.
3:07 The most ambitious, open-ended work goes to Fable. Well-scoped work where speed or cost matters goes to Sonnet. And high-volume, cost-sensitive work goes to Haiku. To measure how each model does on your own tasks and projects, follow the steps in the guide linked below.