Which Claude Model and Effort Level Should You Use for Claude Design Animations?

For most Claude Design animations, start with Claude Sonnet 5 at its default high effort. Move up to Claude Opus 4.8 when the timing is complex or the brief is vague, and raise effort before you reach for a bigger model if the animation looks right but stutters or skips a beat. On July 7, 2026 Anthropic published its own guidance on choosing between the two, which is the first time these controls have had an official decision framework. This piece maps that framework onto the specific job of building an animation good enough to export.

Why Does Model and Effort Selection Matter for Claude Design?

Claude Design's canvas has no model picker and no effort control. You get whatever Anthropic wires up behind the tool, and no dial to turn. Both dials live in Claude Code, which can write and iterate the same React animation code Claude Design produces, and which our Fable 5 and loops guides already use to drive Design projects through a connected MCP server.

The two dials do different jobs. The model setting is roughly how capable the assistant is. The effort setting is roughly how thorough it is on a given task. Anthropic's July 7 post frames the models as people: "Fable is a specialist who's seen problems almost no one else has, Opus is the expert, and Sonnet is a really good generalist." Effort, separately, controls how much reasoning the model spends before it hands work back, including how many times it re-checks its own output.

For animation work that distinction is the whole game. A gradient that loops with a visible jump is rarely a knowledge problem. The model knew how to do it. It just did not check the loop point. That is an effort call, not a model call, and reaching for a more expensive model first wastes tokens on the wrong fix.

What Are the Effort Levels, and Which Is the Default?

Claude Code offers five effort levels: low, medium, high, xhigh, and max. The default is high on Sonnet 5, Opus 4.8, and Fable 5. You set it with the /effort command in a session, or with --effort at launch.

Here is what each level buys you, phrased for animation building:

xhigh and max are available on Sonnet 5, Opus 4.8, and Fable 5. Claude Code also exposes an ultracode option in the /effort menu. It is not a model effort level: it runs at xhigh and additionally lets Claude Code launch multi-agent workflows, which is useful for a big build split across many parts but overkill for a single animation.

Which Claude Model Should Build the Animation?

Match the model to how hard and how vague the animation is, not to how important the project feels. The general rule from Anthropic's guidance: smaller models for changes you can describe precisely, larger models for genuinely hard or ambiguous work.

Sonnet 5 handles the majority of animations people actually make. A confetti effect, an animated background, a title card: these are precise instructions, and a generalist model lands them at a fraction of the cost. Reach for Opus 4.8 when the brief has ambiguity the model has to resolve, like "make the reveal feel snappy but not rushed," or when six elements have to enter in a coordinated sequence. Fable 5 earns its higher price only on the genuinely difficult builds, the ones you would otherwise regenerate five times to get the sequencing right. We cover that trade-off in depth in Building Claude Design Videos With Fable 5.

Model or Effort: How Do You Know Which Dial to Turn?

Diagnose the failure, then turn the matching dial. When an animation comes back wrong, the reason is almost always one of two things, and each points at a different control.

• It got the concept wrong, or produced something confidently off-base. That is a knowledge and capability gap. Move up a model: Sonnet 5 to Opus 4.8, or Opus 4.8 to Fable 5.
• The concept is right but a detail is off: a frame stutters, an element enters late, the loop does not close cleanly. That is a thoroughness gap. Raise effort first: high to xhigh. The model knew what to do and skipped a check.

This order matters for your bill. Effort is cheaper to raise than a model is to upgrade, and animation bugs skew heavily toward the thoroughness kind. Try one step up in effort before you assume you need a smarter model. Most stutters and timing slips clear up at xhigh on the same model that produced them.

How Do You Set the Model and Effort, Then Export?

The dials live in Claude Code, and the export step is the same as always. Five steps from a blank session to a downloadable file:

1. Connect Claude Design to Claude Code, once. Run claude mcp add --scope user --transport http claude-design https://api.anthropic.com/v1/design/mcp, then /design-login to authenticate. This is the same connection our other Claude Code guides use.
2. Pick the model. Run /model and choose Sonnet 5, Opus 4.8, or Fable 5 based on the table above.
3. Set the effort. Run /effort and pick a level, or leave it at high and adjust only if the first result needs it.
4. Build and review. Describe the animation, let it build through the connected MCP server, and watch the result in the browser. If it stutters, raise effort. If the concept is wrong, raise the model.
5. Export to MP4. Claude Design still has no save-as-video button, whatever model built it. Grab the share link and run it through claude2video, which captures the animation frame-by-frame in a headless browser and hands back a clean file. The domain-replace trick still works: swap claude.ai for claude2video.com in the URL bar and it parses automatically.

Frame-by-frame capture is why the model and effort choice pays off at export time. A clean, final animation renders to a clean video with no color loss and no stutter, unlike a screen recording. The less you patched the animation to get it right, the better the final export looks. The full export walkthrough is in How to Export Claude Design Animations as MP4.

Frequently Asked Questions

Can you pick a Claude model or effort level inside Claude Design? No. The Claude Design canvas has no model picker and no effort control. Both dials live in Claude Code, which can build and edit the same React animations Claude Design produces. Connect the Claude Design MCP server to Claude Code, then use /model and /effort to choose. Claude Design itself runs on whatever model Anthropic has wired up behind the tool.

What is the difference between model and effort in Claude Code? The model setting is how capable the assistant is; the effort setting is how thorough it is on a given task. A bigger model knows more and handles harder, vaguer problems. Higher effort makes the same model reason longer and re-check its own work before handing it back. For animations, model gaps show up as wrong concepts, and effort gaps show up as timing bugs like stutters or late entrances.

Which effort level is best for building an animation? Start at high, the default on Sonnet 5, Opus 4.8, and Fable 5. It handles most animations with more than one moving part. Step up to xhigh for a choreographed multi-scene sequence with synced timing, and only reach for max on a genuinely hard visual effect. Drop to medium or low for a single simple element or a one-line tweak, where the extra reasoning is wasted.

Is Opus 4.8 or Fable 5 worth it for a Claude Design animation? Only for genuinely complex builds. Sonnet 5 handles most everyday animations at a fraction of the cost. Opus 4.8 is worth it when the brief is ambiguous or many elements need coordinated timing. Fable 5 earns its higher price on the hardest, most novel effects, where fewer regenerations on an expensive model can cost less in total than many rounds on a cheap one.

Does the model that built the animation change how you export it? No. Whatever model built the animation, Claude Design produces the same output: live React code reachable through a share link, with no native video export. To get an MP4, run the share link through claude2video, which renders the animation frame-by-frame and returns a downloadable file. The export step is identical for Sonnet 5, Opus 4.8, and Fable 5.