Anthropic launches Claude Opus 5.5, gets cost, efficiency memo

Anthropic launched Claude Opus 5.5 and gave a nod to customers who were not happy with their AI costs. According to Anthropic, Claude Opus 5.5 costs 40% less to run than Opus 5.

Yes, Anthropic said Opus 5.5, the first in the Claude 5.5 family of models, performs at a Claude Fable 5.1 level, but performance isn't the big benchmark anymore. Price is the most important thing.

This point has been hammered home repeatedly and Anthropic is feeling heat from not only rival OpenAI, but open weight models and even Nvidia Nemotron. Toss in enterprise data and AI sovereignty and Anthropic had to make its case with Opus 5.5.

The backdrop:

• Nvidia’s Nemotron hit enterprise AI inflection point
• Why you should arbitrage your AI models like your vendors do
• Nvidia’s $12.93 billion Hugging Face acquisition validates open weight models
• Open-Weight Models Are Gaining Ground in Enterprise AI
• Get ready for US open LLMs and just in time
• Rightsizing open models may cut your AI inference spend
• The most important LLM benchmark today: Price

Anthropic made the following points about Opus 5.5:

• Opus 5.5 requires less compute and with default settings will cost about 40% less than Opus 5. Input tokens are $4 per million and output is $20 per million. That price tag is 20% lower than Opus 5. Cache reads are 60% cheaper than Opus 5.
• The model offers better performance in code migrations, one-shot deliverables and accuracy.
• Opus 5.5 generates output 30% faster than Opus 5.
• Opus 5.5 has the best scores on alignment across the Claude portfolio.
• Communication with Opus 5.5 is more natural with clearer writing.

Anthropic said it will follow up with Claude Sonnet 5.5 and Claude Haiku 5.5 and have the same performance and efficiency gains.

Opus 5.5 won't be your most cost effective option, but for use cases that require frontier models Anthropic is giving you a solid option. The main takeaway here is that Anthropic got the message that its models need to be more efficient and competitive on price.

What remains to be seen is whether SpaceXAI and Meta’s Muse efforts that undercut Anthropic and OpenAI on price make a dent. We already know open weight models are here to stay in the enterprise whether it’s via customers that customize models or via your SaaS provider.