Anthropic slashes AI API costs by 75%, and Scottish SMEs should pay attention

Claude's latest model just got dramatically cheaper to run, with cache read prices cut by three quarters overnight. For any Scottish business that uses AI in its workflows, this is a direct reduction in running costs. Here's what changed, and what you can do with it today.

Anthropic has released Claude Fable 5.1 and Mythos 5.1, cutting cache read prices by 75% compared to the previous generation. The two models are technically the same underlying system, separated by their safety configurations: Fable 5.1 is the general-release version, while Mythos 5.1 carries additional safeguards for higher-risk deployment contexts. On the SWE-bench scientific research benchmark, the new model scores 52.6%, against 24.7% for its predecessor. That is not incremental improvement. That is a different league.

The price cut matters more than it might first appear. Cache reads are what happen when an AI model refers back to context it has already processed: previous messages in a conversation, a document you uploaded earlier, a set of instructions you feed in at the start of every session. If you are running Claude regularly across a business, those reads stack up fast. A 75% reduction in that cost changes the economics of building AI into daily operations, particularly for small teams working at scale.

According to research from McKinsey's 2024 State of AI report, the businesses seeing the greatest return from AI adoption are those integrating it into repeated, high-volume workflows rather than using it occasionally for one-off tasks. That is exactly the use case where cheaper cache reads make the most difference. A Scottish accountancy firm running client document analysis, an Edinburgh recruitment agency processing CVs, a healthcare admin team summarising patient correspondence: these are the operations where cost-per-query actually shows up on a balance sheet.

The release also carries a significant regulatory development. Claude Fable 5.1's outputs are watermarked in line with EU AI Act requirements, and Anthropic has opened a private preview of its watermark detection API for eligible organisations. The EU AI Act, which is being phased in through 2025 and 2026, places content provenance obligations on AI-generated material in a growing number of contexts. According to the European Parliament's published implementation timeline, high-risk AI applications face compliance deadlines beginning in August 2026. Any Scottish business operating in European markets, or handling regulated content, should be watching this space now rather than scrambling later.

The practical window here is straightforward. Claude is available via Anthropic's API and through a growing number of third-party platforms including Poe, Amazon Bedrock, and Google Cloud Vertex AI. The Scottish Government's Digital Strategy for Scotland explicitly identifies AI adoption as a priority for SME productivity, and Business Gateway Scotland offers funded digital advisory sessions for businesses at exactly this stage of the journey. If you have been running AI tools in your business but pulling back because API costs were creeping up, the price floor just dropped. That changes the calculation.

What this means for you: If you are using Claude via API, check your billing tier now: the cache read price reduction applies automatically and could cut a meaningful chunk of your monthly AI running costs. If you have not yet built Claude into a repeatable business workflow, such as document summarising, client comms drafting, or data extraction, this is a good moment to start, because the cost of doing so just fell sharply. And if you sell into EU markets or handle regulated content, look up the EU AI Act timeline at europarl.europa.eu and make sure someone in your business owns the compliance question before 2026.