Anthropic released Claude Fable 5 yesterday. Their most powerful model, public for the first time.
Fable 5 is powerful and expensive, and it keeps your data for 30 days. When does thinking cost more than the output is worth?
What it is. One model, two tiers. Fable 5 has safety classifiers for everyone. Mythos 5 is the same model unrestricted, locked to government partners.
What it does. Stripe migrated a 50M-line codebase in a day. Would've taken a team two months. First model to score 90% on Hex's analytics benchmark. One-shots full apps. Works autonomously for days.
The price though... $10/M input, $50/M output. Double Opus 4.8. A complex session with sub-agents and validation loops burns through thousands fast. At what point does the thinking cost more than the output is worth? Eye-bleeding territory.
The data retention catch. 30-day mandatory retention on ALL traffic. Overrides existing zero-retention agreements. The best model also keeps your data for a month. That's a new precedent.
The meta. Anthropic warned about recursive self-improvement last week. Then released their most capable model publicly. Both OpenAI and Anthropic have filed for IPO. The safety messaging and the growth pressure are on a collision course.
Bottom line: genuinely the most capable model you can use. Just keep an eye on the bill and know what you're trading for it.