Cette section n'existe pas encore en français — le contenu ci-dessous est en anglais. La traduction arrive.
Claude Fable 5
Anthropic's highest-tier generally available model, featuring deep reasoning and unusually precise instruction-following.
Quality
Modality
multimodal
Context
1M tokens
Access
closed
Fabian's Take
"Working with Fable is a genuine pleasure. It has a different 'personality' than other models — almost mind-reading when it comes to figuring out what I want — and it holds a whole complex problem without dropping a detail. It was pulled by a US government order for a few weeks in June, but it's back now, and I save it for high-value work: audits, planning, and deep concept work."
Claude Fable 5 is a clear step up from the Claude 4 Opus tier. Rather than just offering raw intelligence, it reads the implicit intent behind complex instructions, works autonomously for long stretches, and communicates in a concise, effective style that wastes little of your attention.
Regulatory pause and return
Shortly after its June 2026 launch, Anthropic suspended access to Fable 5 worldwide under a US government export-control order. The order was lifted after nearly three weeks, and the model returned to general availability on the first of July with new safety safeguards.
How it compares
Fable is much more expensive than Sonnet, but for complex, high-stakes tasks where getting the architecture right the first time matters, the cost is easily justified. Its ability to work independently on a hard prompt without constant course-correction sets it apart from earlier-generation models. For everyday work, though, the standard Claude models are the smarter pick per token.
The Verdict
Best for: Deep concept work, auditing, and planning highly complex projects.
Pros
- Incredible ability to anticipate user needs
- Concise, precise communication style
- Highly capable of working independently on complex tasks
Cons
- Very expensive API pricing — double the Opus tier
- Availability proved fragile — a US government order pulled it worldwide for nearly three weeks from June into July 2026
- Safety filters can over-block legitimate work, and it requires 30-day data retention
Specs
- Pricing $10/M input, $50/M output
- Cost Tier premium
- ⚡Speed Tier moderate
- License Proprietary
