None
None
Long Story Video Skill
Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.
Ads Video Skill
Generate professional ads and sales videos—AI auto-generates scripts, prompts, and footage.
3D Science Explainer Video Skill
Convert scientific concepts into stunning 3D explain animations
Feedback
😢 Every free AI generation costs us real GPU money. We pay out of pocket to keep this free for you.
💎 Upgrade to Pro and unlock:
- ✓HD Download
- ✓No Watermark
- ✓Unlimited Generations
claude fable
Claude Fable 5.2 is the step-up model for tough reasoning and long agentic runs, pairing a 1M-token window with adaptive thinking you steer by effort.
All Tools
Discover our comprehensive AI-powered animation toolkit
Why Build on Claude Fable 5.2
A step-up model for tough reasoning and long agentic runs, Claude Fable 5.2 shares its spec sheet with Claude Mythos 5.2.
- Built for Demanding ReasoningWhen higher-effort evals on Claude Opus 5.5 still fall short, Claude Fable 5.2 is the next stop for hard reasoning and long-running agentic tasks.
- 1M Context, 128k OutputBoth Claude Fable 5.2 and Claude Mythos 5.2 work with up to a million tokens of context at default and maximum, and can emit up to 128k tokens.
- Adaptive Thinking Tuned by EffortThinking can't be switched off; instead the effort parameter sets its depth across low, medium, high, xhigh, and max, with high as the API default.
How to Adopt Claude Fable 5.2
Pick how you build, prove the need with evals, then clear the breaking changes before production.
Claude Fable 5.2 Features
Compaction on demand, mid-conversation tool changes, the computer use toolset, fast mode, strict tool use, and refusal handling on Claude Fable 5.2.
Compaction on Demand (Beta)
The compact-2026-09-04 beta header lets you pick the moment older turns get summarized, so your history stays append-only for the prefix check.
Mid-Conversation Tool Changes (Beta)
Instead of editing the tools array, add or remove tools in a system message; on the Claude API, Claude Platform on AWS, and Anthropic-managed Google Cloud, send the inline-tools-2026-09-15 header.
Computer Use via the Toolset
Reach computer use through the computer_toolset_20260801 toolset; declaring the older computer_20251124 tool now fails with a 400 invalid_request_error, except on Amazon Bedrock.
Fast Mode Availability
Fast mode runs on the Claude API, on both Anthropic-managed and Google-managed Google Cloud, and in Claude in Microsoft Foundry as its own model, claude-fable-5-2-fast.
Schema-Valid Output With Strict Tool Use
For schema-valid JSON, leave tool_choice at auto and add strict: true or switch to JSON outputs; forced tool use errors out, and customer-managed encryption keys rule out structured outputs.
Refusals, Fallback, and Billing
A refusal comes back as HTTP 200 carrying stop_reason refusal and a stop_details object; eligible categories can retry on Claude Opus 4.8 or Claude Opus 5, and fallback credit returns the prompt-cache cost.
Claude Fable 5.2 — FAQ
Common questions about starting, migrating, and pricing Claude Fable 5.2.
Should I start with Claude Opus 5.5 or Claude Fable 5.2?
Start on Claude Opus 5.5 for most workloads; reach for Claude Fable 5.2 only once higher-effort evals on Opus still come up short of what you need.
Why does editing an earlier turn return a 400 error?
Because of the prefix check: it looks at the system prompt, tools, and earlier messages sitting before a thinking block, and any change fails the request. The thinking-binding-controls-2026-08-01 header with prefix_mismatch_behavior set to drop_block is a stopgap.
Which models can read Claude Fable 5.2 thinking blocks?
Only Claude Fable 5.2 and Claude Mythos 5.2 can read each other's blocks. Anything the active model can't read is dropped before delivery, and you aren't billed for those dropped blocks.
Can I turn thinking off or force a specific tool?
No. Sending thinking enabled with budget_tokens, or thinking disabled, both come back 400, and so does tool_choice set to any or to a named tool. Effort levels and strict tool use are the supported routes.
How does Claude Fable 5.2 pricing compare with Claude Fable 5.1?
Both input and output are cheaper than Claude Fable 5.1, which sat at $10 per million input and $50 per million output tokens. Cache reads stay at 0.025 times base input, and batch runs cost $4 per million input tokens.
What happens when a request is refused?
You receive HTTP 200 with stop_reason refusal plus a stop_details object naming the policy area. Some categories can retry via server-side fallback, and billing depends on the category.
Test Claude Fable 5.2 Before You Switch
Prove the need first: check whether Claude Opus 5.5 already satisfies you, settle on a build path, then exercise the four breaking changes on your own integration before production traffic moves to Claude Fable 5.2.