Anthropic has now launched Claude Opus 4.5, the latest version of the company’s flagship AI. Techcrunch writes that Opus 4.5 should perform superbly in several benchmark tests, such as SWE-Bench (coding), tau2-bench (tool usage) and GPQA Diamond (problem solving). It is the first model to score over 80 per cent on SWE-Bench Verified, an important benchmark of a model’s programming ability.
New features include Claude for Excel, a sidebar in the programme now available to Max, Team and Enterprise users. It supports pivot tables, charts and file uploads. At the same time, Claude for Chrome will also be available to all Max users.
Another new feature is improved memory management. Users can now talk to Claude without interruption when the memory limit is reached, by the model itself compressing older parts of the conversation in the background.
Opus 4.5 is also optimised for so-called agentic use cases, where it can act as the main agent and control smaller Haiku-powered sub-agents. According to Anthropic, Opus 4.5 is also their most secure model yet with better protection against prompt injection attacks.
