Anthropic cuts Opus 5.5 costs by 40% with caching and fewer tokens
The savings come from 20% lower input/output prices, 60% cheaper cache reads, and fewer tokens needed per task due to improved efficiency. Anthropic’s blog explains that cache effectiveness—ranging from 5% to 96%—drastically reduces costs, with $11.20 spent on a task when cache is ineffective versus $0.99 when it is 96% effective. For comparison, Claude Fable 5.1 costs $10 per million input…
Key points
- Opus 5.5 costs **40% less** than Opus 5 for typical tasks due to **lower unit prices** and **fewer tokens used**
- Cache effectiveness cuts costs from **$11.20** to **$0.99** for the same task, depending on how well it works
- Anthropic recommends **/compact**, **medium effort**, and **no mid-session model switches** to maximize savings
Model pages: Claude Opus 5.5 → · Claude Fable 5.1 →
How much cheaper is Claude Opus 5.5 really? Decoding the official calculations in English (with Claude Code settings)
note.com · 24 September 2026
Loading the full article…
This text was published by note.com and written by なぎ. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Anthropic engineer says Claude's writing declined as newer models focus on code · 2 src
- AI expert warns machine-speed cyberthreats may outpace human defenses · 1 src
- OpenAI may unveil GPT-6 Cyber cybersecurity model in weeks · 5 src
- Fastino Releases GLiNER2.5-Decide: A 340M Open-Weight Decision Model That Runs on CPU · 1 src
- Pistis introduces 27B- and 9B-parameter multimodal models via new training framework · 1 src
Comments
via GitHub Discussions