Credit: Anthropic Anthropic released Claude Opus 5.5 on Tuesday at $4 per million input tokens and $20 per million output, below the $5 and $25 that Opus 5 has carried since July, and says it performs similarly to the more expensive Fable 5.1 on most work. It is the first model since Dario Amodei called for third-party evaluation, which Article 55 of the AI Act has obliged in Europe since August 2025. Anthropic released Claude Opus 5.5 on Tuesday, a model it says performs similarly to its more capable Fable 5.1 on most work while costing significantly less to run than Opus 5, Bloomberg reported. Sonnet and Haiku versions follow in the coming weeks.
It is built for long-running agentic coding and knowledge work, with a one million token context window. The price is the news. Opus 5.5 costs $4 per million input tokens and $20 per million output, down from $5 and $25. Fable 5.1 is $10 and $50.
Cache reads are $0.20 and batch processing is half price. That is the first cut to an Opus price. When Opus 5 arrived in July it carried the same price as the model before it. Anthropic has spent a year moving capability without moving the number.
This release moves the number instead. The other half is evaluation. Anthropic says METR and Frontier Design tested Opus 5.5 before release, and that it carries safeguards like Fable’s to limit riskier cybersecurity and biology uses. The model now runs a biology classifier alongside the cybersecurity one.
Adaptive thinking cannot be switched off, and the default effort level drops from high to medium. Ten days ago that was a promise. Dario Amodei called on the industry to slow the pace of AI development on 12 September, and said his company would bring in third-party evaluators. This is the first model since.
Europe has required it since August 2025. Article 55 of the AI Act obliges providers of general-purpose models with systemic risk to run adversarial testing to standardised protocols, assess and mitigate systemic risk, protect the model and its infrastructure, and report serious incidents without delay. ENISA already evaluates models including Anthropic’s. The AI Omnibus left those obligations unchanged in July.
So the step is voluntary in one market and compulsory in another. The price has a reason too. Customers have been moving to cheaper models, and Harvey, the legal startup valued at $15.6B, built its own on a Chinese open-weight base in August to displace what it was renting. Anthropic lists on Nasdaq next month, having filed confidentially at $965B.
Leave a Reply