AI Pulse · Frontier Labs Watch

Anthropic has cut cache-read pricing for Fable 5.1 by 75%, from $1.00 to $0.25 per million tokens. On paper, that can reduce typical workload costs by up to 45%. That’s a real input-cost reduction, and it matters for teams running repeat prompts, large knowledge bases, or workflows that reuse lots of context.

But the number your business feels is cost per completed task. Artificial Analysis’ pricing data puts Fable 5.1 at maximum effort at $3.76 per task, compared with $3.14 for Fable 5. That is about 20% more. The reason is simple. The new model uses roughly 1.7 times as many output tokens, and the saving on cached input does not fully cover the extra generation cost. That helps explain reports from users hitting Claude Max limits far faster than expected. It’s not necessarily user error. It’s the difference between a lower price on one part of the bill and a higher total bill for the work being done. Artificial Analysis’ pricing data

For owners, the practical lesson is to test model changes against your own jobs, not vendor headline pricing. Track cost per proposal drafted, support case resolved, report produced, or workflow completed. Also watch the quieter updates. Anthropic has restored customer-controlled zero data retention through its Enterprise Frontier Safeguards, and added EU AI Act watermarking with a regulator detection API in private preview. This is the kind of thing we build into an AI command centre, where usage, cost, governance, and model performance are visible in one place.