Anthropic It recently launched two new artificial intelligence models in the series Claude 4that is to say Opus 4 And Sonnet 4. Although they share many technological innovations, the two models have been designed with different purposes and areas of use. Claude Opus 4 It represents the absolute summit in terms of power and deep reasoning capacity, designed to face very complex activities and long -lasting projects with unprecedented precision. Claude Sonnet 4Instead, it has been optimized for efficiency and accessibility, offering high performance in practical areas of everyday use, with quick response times and understanding of the context improved with respect to its predecessor, the Sonnet 3.7. Both models integrate one new function of “extended thought” and are able to use external toolslike web research, during reasoning processes. In addition, they boast an advanced memory that allows them to retain relevant information over time. Claude Sonnet 4 is also available for free users, while Opus 4 is included in the most advanced paid plans.
The most interesting features of the new Claude 4 models
One of the most interesting innovations introduced in both models is the so -called “Thought extended with the use of tools”. In practice, when a request requires it, the model can alternate moments of internal reasoning with the use of external tools – how to access information on the Internet – to build a more complete response. This ability, combined with one enhanced memory (which, if authorized, allows the model to save and recover information between different sessions), allows Claude to face complex scenarios with a continuity that brings it closer to human behavior. Regarding this last aspect, Anthropic said:
When developers create applications that provide access to Claude’s local files, Opus 4 becomes skilled in creating and maintaining “memory files” to archive the key information. This unlocks a better awareness of long -term activities, consistency and performance on the activities of the agents, such as Opus 4 that creates a “navigation guide” while playing Pokémon (see the following image).
From the point of view of software development, Claude 4 is also enriched with the system Claude Codea set of tools dedicated to software development that allows you to integrate Claude directly in the terminal, in the main development environments (such as VS Code and Jetbrains), and to use it in secondary execution through a SDK.
On the front of Bee (Application Programming Interface), Anthropic has introduced new functions designed to facilitate the creation of advanced intelligent agents. Among these, the code execution, file management, integration through connectors and the possibility of temporarily memorizing prompts to improve their reactivity stand out. All this greatly expands the possibilities of customization and control for those who develop applications based on Claude.
Last But Not Leastshould be underlined theAnthropic’s commitment to the reliability front. The new Claude 4 models, in fact, show a 65% reduction in the use of logical shortcuts or inaccurate responses compared to the previous version. This is a fundamental fact especially for the most sensitive applications, where the correctness of the information is essential.
The differences between the Claude Opus 4 and Sonnet 4 models
To better understand what distinguishes Claude Sonnet 4 from his “older brother” Opus 4we start from the characteristics of the latter. Opus 4 is currently the most advanced model produced by Anthropic, designed to face complex workflows and long -lasting activities with continuity and precision. It stands out above all in the context of software development, where it has overcome all the other models in the benchmark Swe-Bench (72.5%) e Terminal-Bench (43.2%). These tests evaluate the effectiveness of a model in correcting, generating and understanding real code, activities that require long -term memory, understanding of addictions between files and ability to maintain the context even after many iterations. The performance of Opus 4 were validated by big names in the sector such as Replica, Github And Rakutenwho have praised the ability to manage prolonged sessions without loss of consistency or quality.
In parallel, Claude Sonnet 4 It was developed to bring a part of these innovations to a more accessible and light context, ideal for daily use scenarios. While not reaching the computational peaks of Opus 4, Sonnet 4 manages to combine speed and precision in a balanced way, resulting perfect for developers, analysts and creatives who need a powerful and at the same time reactive tool. He also shows excellent results in the reference benchmarks, as a score of 72.7% on Swe-Benchand proves to be particularly effective in following complex instructions, generating clean and structured code, and perform more articulated reasoning activities than version 3.7.
