None
None
Long Story Video Skill
Let Claude Fable 5.2 orchestrate a 10s–10min story, handling scripts, prompts and footage end to end.
Ads Video Skill
Claude Fable 5.2 drafts the script, prompts and footage for a product ad in a single pass.
3D Science Explainer Video Skill
Turn a hard research result into a 3D explainer animation
Feedback
freeTrialImage.bannerPity
freeTrialImage.upgradeUnlock
- ✓freeTrialImage.benefitHd
- ✓freeTrialImage.benefitWatermark
- ✓freeTrialImage.benefitUnlimited
Claude Fable 5.2
Claude Fable 5.2 pairs code-level judgment with vision and long-horizon reasoning, so heavy engineering and research jobs finish about 30% sooner.
All Tools
Discover our comprehensive AI-powered animation toolkit
MiniMax H3
MiniMax H3 AI Video Generator
Seedance 2.5
The Future of AI Video Is Here.
Seedance 2.0
The Future of AI Video Is Here.
Seedance 2.0
The Future of AI Video Is Here.
Gemini Omni
Gemini Omni Video Generator
Image to Video
Generate videos from images
Gemini Omni 1.2 Flash
AI Multi-Scene Shorts Generator
Create viral AI Shorts instantly
Anthropic's Claude Fable 5.2, Explained for Builders
Anthropic built Claude Fable 5.2 for work that resists a quick answer — moving a live service to a new language, or chipping at questions physics has left open.
- Ship-Ready Coding JudgmentMore than nine in ten merged changes in Anthropic's own codebase now come from Claude. Fable 5.2 cleared long-standing bugs in minutes and ported a core service from Python to Rust, cutting startup time roughly 26×.
- Progress on Unsolved Science and MathWorking through holographic duality, Claude Fable 5.2 closed two unresolved questions in black hole physics, then formalized the proof of Fermat's Last Theorem — a run that took 30 hours.
- Guardrails, Plus Verified Access via MythosFable 5.2 and Mythos 5.2 share a single underlying model. The Fable release ships with cybersecurity and biology guardrails so it stays open to everyone, while security and biomedical teams request Mythos through the Life Sciences and Cyber Verification Programs.
Setting Up Claude Fable 5.2 for a Task
Align three dials — model tier, effort level and mode — with how large and how novel the job in front of you actually is.
Claude Fable 5.2 by the Numbers: Benchmarks and Strengths
Results across FrontierCode, Chartography, WANDR, HLE and BenchCAD, plus multi-day agent orchestration, CVE hunting in Mythos 5.2, and lab hardware control via MHS.
FrontierCode: Merge-Quality Scores
Running at High effort, Claude Fable 5.2 posted 45.3% on FrontierCode 2.0 — level with Opus 5.5's 45.5% best at roughly half the cost per task, and clear of GPT-6 Astra's 42.8%. Bumping to Xhigh lifted it to 49.1%.
Agent Runs Measured in Days
From an empty folder, Fable 5.2 directed more than 1,000 subagents across eight days to produce a bootable operating system — drivers, file storage, windows, installer — along with a hypervisor resting on over 3,500 machine-checked proofs.
Bug Hunting with Mythos 5.2
Against 1,829 CVEs in the internal CVE Rediscovery test, Claude Mythos 5.2 surfaced 10.2% more flaws than Opus 5.5 and 14.4% more than Mythos 5.1. One security firm found that a sixth of the bugs it confirmed had slipped past every other frontier model in the comparison.
Reading Charts and Images
At Medium effort Fable 5.2 hit 66.4% on Chartography, edging Opus 5.5's Max-effort 64.4% at under a tenth of the per-task cost. On BenchCAD 2.1 it landed 83.5 versus 65.3 for Opus 5.5.
Cheaper Research and Reasoning
Fable 5.2 at Medium effort equalled Opus 5.5's best WANDR score while spending about 40% as much per task, and at Low effort it matched the top Humanity's Last Exam result for less than a tenth of the cost.
Driving Lab Hardware through MHS
Under the Model Hardware Standard research preview, one MHS driver built on simple read and write primitives lets agents operate microscopes, liquid handlers and robotic arms side by side — shrinking integration from weeks to hours for any programmable device, over the Model Context Protocol and any agent harness.
Claude Fable 5.2: Common Questions Answered
Pricing, throughput, model selection, safety testing and Mythos access — the questions teams ask most about Claude Fable 5.2.
How do Fable 5.2 and Mythos 5.2 differ?
They are the same underlying model released two ways. Fable 5.2 ships with cyber and biology guardrails so it can be offered openly; Mythos 5.2 is granted to vetted security and biomedical teams.
What does Claude Fable 5.2 charge per token?
Standard rates are $8 per million input tokens and $40 per million output tokens. Fast mode, when enabled, runs $16 per million input and $80 per million output.
Is Claude Fable 5.2 quicker than Fable 5.1?
Output arrives roughly 30% sooner than Fable 5.1, and fast mode can multiply token generation by up to 2.5×.
Should I reach for Fable 5.2 or Opus 5.5?
Default to Opus 5.5 for everyday tasks; escalate to Fable 5.2 when the work is unusually broad, novel or long-running, and step back down when it is not.
Which safety checks ran before launch?
Pre-release evaluations probed sandbox escapes and other risky behaviour, confirmed alignment on par with recent models, and tested leaner safeguards alongside tougher anti-distillation defences.
Is Claude Fable 5.2 cleared for cyber or biomedical work?
Most cybersecurity requests are blocked on Fable 5.2, so that work requires Mythos through the Cyber Verification Program. Claude Security now ships with Mythos 5.2 included.
Hand Your Hardest Problems to Claude Fable 5.2
Start at $8 per million input and $40 per million output tokens, switch on fast mode whenever latency matters, and request Mythos access through the Cyber and Life Sciences Verification Programs.