Claude Fable 5.2 and Mythos 5.2 for Motion Design

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026. If you make motion graphics, launch videos, social content or animated websites, the benchmark table isn't the story. The story is that the best model you can actually use got faster, cheaper and noticeably better at seeing.

This guide covers what changed, which model to use for which job, and how to put the new models to work in a motion editor instead of a chat thread. Everything here is as of November 10, 2026.

What Anthropic released on November 9, 2026

Two names, one model. Anthropic says Fable 5.2 and Mythos 5.2 share the same underlying model. What differs is the safeguards and who gets access.

• Claude Fable 5.2 ships with cybersecurity and biology safeguards and is available on all paid Claude plans, on the Claude Platform (API model ID claude-fable-5-2) and on the major clouds.
• Claude Mythos 5.2 drops some of those safeguards and is limited to vetted organizations through Anthropic's Cyber Verification and Life Sciences Verification programs.
• Price: $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. Fast mode runs up to 2.5x faster at $16 and $80.
• Speed: output up to 30% faster than the 5.1 models, and fewer tokens used at the same effort level.
• Effort: Low, Medium, High, Xhigh and Max. Medium is the default in the Claude apps and Claude Code; High is the default on the API.
• Context: a 1M-token context window and up to 128K output tokens, per Anthropic's model docs.

On Pro plans and standard Team seats, Fable can use up to half of your weekly usage limit. Claude Opus 5.5, released September 22, stays in the lineup at $4 and $20 per million tokens, and Anthropic still recommends starting there for most everyday work.

Do motion designers need Mythos 5.2?

No. Mythos 5.2 is the same model with fewer guardrails around cybersecurity and biology research. Nothing in motion design, animation or video sits behind those guardrails, and access runs through verification programs built for security and life-science teams.

If you saw "Mythos" in a headline and wondered whether you're missing something, you aren't. Fable 5.2 is the one to use.

What changed for motion design, animation and video

Anthropic's announcement is written for coding, finance and science. It doesn't mention animation once. Read it with a motion designer's eye, though, and four things stand out.

It reads images better

On Anthropic's Chartography test, which asks a model to read charts without tools, Fable 5.2 scored 73.4%, up from 44.8% for Fable 5.1 and 64.4% for Opus 5.5. It's a chart benchmark, but the skill underneath is the one you lean on every time you hand a model a reference frame, a storyboard, a screenshot of your app or a competitor's launch video and say "like this."

It builds better 3D from scratch

A Unity tester quoted in the launch post said Fable 5.2 is good at building higher-quality 3D scenes from scratch and at rebuilding an experience from a visual reference. If you prompt for three.js scenes, WebGL shaders or interactive product heroes, that's the line to pay attention to.

It keeps going until the job is done

Anthropic's testers say it "works until the job is actually complete" instead of stopping at a plausible checkpoint. In motion terms, that's the gap between a model that hands you the first 8 seconds of a 30-second launch film as a "draft" and one that builds all three sections and then checks them.

Top-tier output costs less

In Anthropic's early testing on one coding benchmark, Fable 5.2 at High effort scored about level with Opus 5.5 at Max, for roughly half the cost per task. Add the 20% price cut and Fable 5.2 stops being a model you save for emergencies.

One honest caveat: every number above comes from Anthropic, and none of them measures motion design directly. Judge it on your own scenes.

Opus 5.5 is where prompting alone started to work

Fable 5.2 lands on top of a model that already moved the bar. With Opus 5.5 in September, one well-written prompt could return a complete, good-looking animation: a launch title timed with GSAP, a scroll-driven section, a three.js product scene with believable light. When testers had several Claude models build a game from a single prompt, Anthropic reports Opus 5.5 scored highest for graphics and polish.

That's why Opus 5.5 is the default model in Paraplex's AI assistant. It's fast, it costs half as much as Fable 5.2, and for most scene building and edits it's enough.

Where prompting alone stalls

Here's the part launch posts skip. Getting a first draft out of a model is now the easy part. The revision loop is where the hours go.

• Every tweak is a full round trip. "Move the title up 40 px and ease it out slower" means asking, waiting for a regenerated file, rendering it again and hoping nothing else changed.
• You get code, not a project. To nudge one keyframe yourself, you edit JavaScript. Most designers don't want to, and shouldn't have to.
• It isn't on brand by default. Fonts, colors, materials and motion rules have to be restated in every prompt.
• It's one format. The 16:9 cut doesn't become a 9:16 Reel without another round of prompting.
• It doesn't ship itself. Turning an HTML file into an MP4, an embed on your site or 500 personalized versions is a separate project.

A better prompt won't fix this. Putting the model's output somewhere you can edit it directly will, with the busy work automated around it. That's the job Paraplex was built for: the model does the first draft and the tedious parts, and everything it makes stays an editable layer, keyframe or control.

How to use Claude Fable 5.2 and Opus 5.5 in Paraplex

There are four routes, and they stack. Most people start with the assistant, paste code when a model writes something they like, and move to Claude Code or Automations once a format starts repeating.

1. Ask the built-in assistant

Paraplex's AI assistant builds and edits scenes with 158 tools covering layers, keyframes, camera moves, materials, lights and charts. It can build a whole scene from a sentence, or change one element you point at by clicking a layer or dragging a box over part of the frame. Every turn is a preview you Apply or Discard, and one undo reverts the whole turn.

For anything longer than a title, turn on Director mode. The assistant writes a plan and a storyboard of stills for your approval, then builds the piece section by section. It renders up to six moments to check its own work and fixes what it finds, like a title running past the edge of the frame.

Pick the model in the composer. Claude Opus 5.5 is the default, and Claude Fable 5.2 is there for the hard builds, with effort from Low to Max. Add your own Anthropic API key in the Account window and the assistant runs on your key instead of Paraplex credits. Keys stay in your browser and are never saved in projects.

The real value is what the assistant takes off your plate:

• Keyframes. It sets timing and easing across every layer, and you can still grab any keyframe and drag it.
• Music. Beat sync detects the beats in a track, clips snap to them, and the assistant can key a setting to hit on every beat.
• Brand. Save your colors, type, materials and motion as a design system and restyle any scene onto it in one step.
• Formats. Dynamic artboards frame one scene for 16:9, 9:16, 1:1 and 4:5, each laid out again for its shape.
• Type. 40+ text animation presets that work on any text layer, 2D or 3D.

After that, it's yours. Every layer, keyframe, camera move, material and shader stays editable by hand. For the longer walkthrough, see AI motion design in Paraplex and text to animation.

2. Paste code from Claude, ChatGPT or Gemini

If you already prompt for animations in a chat app, keep doing it. Paste the result into Paraplex and it becomes a live code layer on the timeline: HTML, CSS and JavaScript, a React component or a shadcn/ui command. It runs on the timeline's clock, so it exports frame for frame instead of being screen-recorded. Code from any model works.

That includes three.js. import * as THREE from "three" loads a pinned build, and React Three Fiber and drei scenes run too, so the WebGL scene a model writes for you plays live on the timeline and exports like any other layer.

Then Paraplex gives the code controls you can see:

• CSS variables become dials. Each one is a keyframeable control, so you change speed, color or size with a slider instead of editing the file.
• Shaders get controls too. Write or paste a GLSL fill, a full-frame effect or a Shadertoy, and every number, color or on-off parameter becomes a keyframeable control with a min, max and default. The Shader panel tells you whether it compiles and points errors to the line.
• Or describe it. Ask the assistant for a generator, a small JavaScript program, and it writes one with its settings exposed as controls.
• Lift it into layers when you'd rather finish by hand, or keep it live.

The same piece can ship as an interactive embed. Publish embed gives you a two-line snippet that places the live scene on any site, where it answers hover, click, drag and scroll and its code layers keep running. That's the route for website hero animations, scroll-driven sections and interactive product scenes. More on the code side in HTML to video.

3. Let Claude Code drive Paraplex

If you work in Claude Code, connect it to Paraplex with the Paraplex MCP server: claude mcp add paraplex -- npx -y paraplex-mcp. The agent attaches to your open Paraplex tab, builds scenes with the same tools as the assistant, looks at frames to check its work, and exports. Each edit is one undo step, so you can watch it work and take back anything you don't like.

This is where Fable 5.2's long runs earn their keep. Fast mode is available in Claude Code, so an agent session that builds a whole launch film doesn't have to mean a long wait. Setup is in the Claude Code guide, and the bigger picture is on AI agents for motion design.

4. Automate it with Automations, the Render API and n8n

The last step takes you out of the loop for work that repeats. Automations, available on every plan, render your templates without you opening the editor.

• Starts: by hand, from a webhook, on a schedule (hourly, daily or weekly), from an RSS or Atom feed, or from each new row in a Google Sheet or CSV.
• Steps: Fetch, Ask Claude, Render, Wait, Ask me to approve, Call a webhook, Email me, Send to Slack, Upload to storage and Post to social, plus If, For each and Stop. Up to 50 steps per automation.
• Checks: a Render step can flag content past the frame edge, text under 3:1 contrast and type under 14 px, and hold the run until you approve it.
• Delivery: Slack, S3-compatible storage, or social posts through a Postiz connection to X, Bluesky, Facebook, Instagram, LinkedIn, Pinterest, Reddit, Threads, TikTok and YouTube.

Developers can skip the builder. The Render API starts, follows, cancels and lists renders, fills up to 64 template slots per render, applies a design system and calls your webhook when the MP4 or WebM is ready.

If your workflows live in n8n, the open-source Paraplex node (n8n-nodes-paraplex) creates, gets and cancels renders, looks up templates, runs an automation and checks its runs, with triggers for Render Finished, Render Failed and Automation Run Ended. Make and Zapier integrations are coming soon.

Put together, that's fully automated social video: a feed or sheet starts the run, Claude writes the copy, Paraplex renders every format, and a person approves before anything posts. If your content already lives in a spreadsheet, data-driven video links layers to CSV or JSON columns and makes one video per row.

Example: a personalized video for every form submission

This is the workflow people ask about most. Someone fills in the demo form on your site, and a few minutes later they get a short video with their name and company in it, on brand, checked and delivered.

1. Build the template once. Design the video in Paraplex, by hand or with the assistant, and mark the text and pictures that change as slots: name, company, a logo, a one-line opener.
2. Create an Automation that starts from a webhook. Paraplex gives it a webhook address, and you can make a new one any time.
3. Point your form at it. Any form tool that can send a webhook works, and so does an n8n workflow in between. On a Webflow site, a form submission webhook does the job (see Paraplex for Webflow).
4. Add an Ask Claude step. It turns the form answers into a one-line opener, so the copy changes with the person, not only the name.
5. Add a Render step. It fills the slots, renders MP4 or WebM and checks the frames. Keep the review hold on for the first few runs if you want to see each video before it goes out.
6. Deliver it. Call a webhook into your email tool or CRM to send it to the lead, post it to Slack for your sales team, or upload it to your own storage.

Swap the trigger and the same template makes a weekly stats recap on a schedule, a video for every new blog post from your RSS feed, or a welcome video for every new customer row in a sheet. It's the bridge between launch videos you make once and social videos you need every week.

Which Claude model to use for which motion job

• Everyday scenes and edits: Opus 5.5. Titles, product reveals, chart animations, restyles and "make it warmer" changes. Fast, half the price, and Anthropic's own recommended starting point.
• Rebuilding from a reference: Fable 5.2. When you hand over a screenshot, a storyboard frame or a competitor's spot and want it matched, the vision gains show up here.
• Multi-section films in Director mode: Fable 5.2. Long plans that need every section built and checked.
• Heavy code: Fable 5.2. Shaders, WebGL scenes and interactive embeds, where one wrong line breaks the frame.
• Long Claude Code sessions: Fable 5.2, with fast mode if waiting costs you more than tokens do.

Effort matters as much as the model. Medium is right for most turns. Go to High for a complex first build, and save Max for the moments you'd otherwise spend an hour fixing by hand.

What this means for motion designers

Every new model makes the first draft better. Fable 5.2 makes it better at seeing and at building 3D. None of that removes the part of the job that decides whether the work is good: the taste calls, the timing, the hundred small edits between "close" and "done."

Prompting and re-prompting until a file looks right is fine for a one-off. For work you'll revise, reuse and ship in every format, let the model build the scene and handle the keyframes, the beat sync, the brand and the formats. Then finish it by hand, and let Automations make the next 500.

Paraplex is free to start in your browser, with 1080p exports and no watermark. See how it compares in the best AI motion design tools, how it differs from Anthropic's own Claude Motion, or check pricing.