After the launch of Opus 4.1 last week, they have been working at a «fast clip,» to compete with Google and OpenAI, The Verge writes.
Claude now comes with memory
First came the announcement of «memory» in Claude on Monday. This is a handy new feature that allows the chatbot to remember things from previous chats.
It is, however, not persistent, like it is for ChatGPT and the later Gemini launch. Instead, you need to invoke it by asking a question, like «what were we working on last week?»
Claude will then skim through your past chats and come up with a comprehensive answer.
There is no way to exclude a chat from the memory feature aside from deleting it, or by turning off «Search and reference chats» in Settings under Profile.
1 million context window
Then, on Tuesday, Anthropic launched what might well be a game changer: Claude Sonnet 4 now supports 1M tokens in the API interface.
That’s a 5x increase to the previous max limit, and lets you «process entire codebases with over 75,000 lines of code,» or «dozens of research papers» in a single request.
1M tokens also translates to roughly 750,000 words, which is more than the entirety of Lord of the Rings by Tolkien — that you can process in a single prompt.
The context window for the recent GPT-5 Thinking is 196,000 tokens, and this move finally catches up with Gemini 2.5 Pro.
— We’re also exploring how to bring long context to other Claude products, Anthropic says.
$1 for every government agency
Then, also on Tuesday, Anthropic decided to match Open AI’s government offer, giving «all three branches of government» access to Claude for a year for a simple dollar.
— We believe the U.S. public sector should have access to the most advanced AI capabilities to tackle complex challenges, from scientific research to constituent services, Anthropic says on their launch page.
This includes Claude for Government for unclassified, but sensitive work, and Claude for Enterprise — both giving access for their «frontier models.»
The program includes technical support and training to «help agencies rapidly implement AI into their productivity and mission workflows.»
So while others might bask in the sunlight, Anthropic have been hard hard at work on coding, not just iterations on new models — but on killer features.
And Claude for government was actually requested:
— We encourage other American AI technology companies to follow OpenAI’s lead and work with us as GSA’s OneGov continues to modernize and streamline government operations, the General Services Administration’s Federal Acquisition Service Commissioner, Josh Gruenbaum, said about a week ago in response to OpenAI’s offer.
Read more: The Verge on memory and context window, TechCrunch on the government plan.