Anthropic: New Claude models have knowledge through June 2026

What's new

Anthropic states in its own model overview that the four current Claude models, Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, all have June 2026 as both their reliable knowledge cutoff and their training data cutoff. The knowledge cutoff is the date up to which the model's knowledge is most comprehensive and reliable. The training data cutoff is the broader span of data used to train it.

For comparison, older models have earlier cutoffs. According to the independent overview from metehan.ai, Fable 5, Sonnet 5, Opus 4.8 and Opus 4.7 have January 2026, Opus 5 has May 2026, Sonnet 4.6 and Opus 4.6 have August 2025, and Haiku 4.5 has July 2025. However, the overview is partly based on a search and has not been confirmed on a page for each model.

What's smart

A knowledge cutoff tells you how far forward in time an AI model "knows" the world without looking anything up. The more recent the cutoff, the lower the risk that it answers based on outdated information about news, technology or regulations.

All four Claude models have a context window of 1 million tokens, meaning an amount of text the size of several thick books that the model can keep track of at one time. They can respond with up to 128,000 tokens. Anthropic also promises not to retire the models before the fall of 2027: Fable 5.1 no earlier than September 1, 2027, Opus 5.5 September 22, Sonnet 5.5 September 28 and Haiku 5.5 October 7, 2027. That gives companies peace of mind when building on them.

Cheaper or better?

By Anthropic's own figures, the Claude models are about six months ahead of OpenAI's GPT-5.5, which has December 1, 2025 as its knowledge cutoff. GPT-5.5 is dated April 23, 2026 and has a context window of 1,050,000 tokens. The cutoffs are the providers' own statements and have not been independently measured. Metehan.ai also warns that models cannot reliably report their own cutoff, so you should check the documentation. The source also points out that a late-released model may well have an old cutoff, for example Google's Gemini 3.x series.

API prices per million tokens (input / output): - Claude Fable 5.1: $10 / $50 - Claude Opus 5.5: $4 / $20 - Claude Sonnet 5.5: $2 / $10 - Claude Haiku 5.5: from $0.10 / $0.50 - GPT-5.5: $5 / $30

Anthropic's Batch API offers a 50 percent discount. At OpenAI, cached input costs $0.5 per million, and prompts over 272,000 input tokens cost double the input price and 1.5 times the output price for the entire session. Opus 5.5 is thus cheaper than GPT-5.5 on both input and output. Monthly prices for regular subscribers and availability in Denmark and the EU are not stated in the sources.

What it's good for

Anthropic recommends Opus 5.5 for most tasks and Fable 5.1 for demanding reasoning and long-running agent work, where the AI works independently through many steps. Haiku 5.5 is the cheapest and suits large volumes of simple tasks, while Sonnet 5.5 sits in between.

OpenAI positions GPT-5.5 as a leading model for the most complex professional tasks. It accepts text and images as input and responds with text.

In practice, a more recent knowledge cutoff is mostly an advantage when you ask about recent events, tools or developments. For tasks where you supply the material yourself, it matters less.