Since its establishment in 2021, Anthropic has quickly become one of the AI companies and is a qualified competitor for Openai, Google and Microsoft with its cloud model. Construction at this pace, the company organized its first developer conference, Thursday, – Code with Cloud – which showed what the company has done so far and where it is going forward.
(Disclosure: ZDNET’s original company Ziff Davis filed a case of April 2025 against Openai, alleging that it violates Ziff Davis copyright training and operating its AI system.)
Also: I let Google’s Jules AI agent go to my code repo and it worked for four hours in a moment.
Anthropic used an event stage to unveil two highly anticipated models, Cloud Opus 4 and Cloud Sonnet 4. Both provide improvement on their predecessor models, including coding and better performance in logic. In addition, the company launched new features and equipment for its models, which should improve user experience.
Keep reading to learn more about new models.
Cloud Opus 4
The Cloud Oppus family has always been the company’s most advanced, intelligent AI model that is leading to complex tasks. While Cloud Oppus 3 was already famous as a highly capable model. The latest generation has made it even more. Anthropic has referred to it as the most powerful model and the best coding model in the world, supported by the results of Swe-Bench, which you can find below.
Anthropic stated that the Opis 4 was designed to give a continuous performance on complex, long -lasting tasks, requiring thousands of steps, which improves all of all the sonnet models better. One of the largest highlights is that the model can run autonomously for several hours, causing a great model to power the cloud Opus 4AI agents – the next marginal of AI aid.
Also: Top 20 AI tools of 2025 – and #1 to remember the cheese when you use them
AI agents appeal lies in their ability to work for people without intervention. To do this successfully, they need to argue through the next required stages, such as which equipment to call or to take which action. As a result, agents require a model that can argue well over time and maintain the argument – such as Cloud Ops 4.
Cloud sonnet 4
As the next generation of the Cloud Sonnet family, Cloud Sonnet 4 maintains the appeal of its predecessor model, a highly capable yet practical model fit for the needs of most people. Cloud Sonnet 4 has been manufactured on the characteristics of Cloud Sonnet 3.7 with better stereability, a word that states how well a model can take human direction, argument and coding. This will now be a drop-in replacement for Cloud Sonnet 3.7 in Chatbot.
Other improvements for clouds
A new feature available in beta allows Opus 4 and Sonnet 4 to alternate between extended thinking and use of equipment, allowing users to experience a composite performance that combines speed with accuracy. Anthropic stated that the cloud can also call the tools in parallel, which means that it can either either sequentially or simultaneously call the function to execute the task.
Too: Ethropic mapped cloud morality. What’s the chatbot value here (and no)
When developers give cloud access to local files, it can now create and maintain a “memory file” with the major insights, which allows for “better long -term work awareness, consistent and performing agent functions” according to anthropic. Developers provide new capabilities in anthropic APIs for the manufacture of more powerful agents, including code execution tools, MCP connectors, file APIs, and one hour supported prompt cashing.
Another improvement in both models is a 65% decrease in reward hacking – a behavior where the model takes a shortcut to complete a task – compared to Cloud Sonnet 3.7, especially on agentic coding works where this issue is common.
The users will also achieve increased insight into the model thinking process with a new thinking summary feature. This feature displays the logic of the model in digestible insights rather than the raw range of thought when thought processes are very long.
Anthropic stated that the summary would require only 5% time, as most of the processes are sufficient to display completely. After insight on how to reach a conclusion, users help users to verify its accuracy, identify any intervals in the process, and perhaps learn how they can come to the answer themselves.
Too: According to Anthropic, college students are using Cloud AI
Anthropic also announced plans for the company’s future, including preparing models for high AI security levels such as ASL -3 and providing more frequent models updates to use rapid success capabilities.
Standard
With any model release, the launch of Opus 4 and Sonnet 4 was with benchmark results. Both models performed extraordinary performances in coding works. When SWE-Bench is verified, a benchmark to evaluate large language models on real-world software challenges, which requires agentic regioning and multi-step code generation, OPUS 4 and Sonnet 4, performed better than many major models in coding domains, in which Openai Codex-1, OPENAI O3, O3, O3, O3.1.1, and Gemini 2.5 Pro includes.
Beyond coding, Ops 4 and Sonnet 4 also performed competitively, either leading or getting closer to it, in other traditionally used benchmarks, including GPQA diamond, which tests for graduate level logic; AIME 2025, which tests high school match competition level; And Mmmlu, which tests for multilingual functions.
Availability
Cloud Ops 4 and Sonnet 4 are hybrid models, with close-instant response mode and an extended logic mode for requests that require intensive analysis. In paid cloud plans, including Pro, Max, Team and Enterprise, both models and extended thinking have access to. Cloud Sonnet 4 is also available to free users.
Developers can use both models on anthropic API, Amazon Bedrock and Vertex AI of Google Cloud. Anthropic shares that correspond to the previous model.
Bonus: Cloud Code
Cloud code developers directly using the coding assistant of the cloud to the developers where they write and manage, whether in the terminal, are running in the background with their IDE, or cloud code within the IDE, or the cloud code. For example, new beta extensions for VS code and jetbrances allow users to integrate cloud code within the ID, where the proposed editing of clouds will appear inline.
Also: I tested the intensive research of chat against Gemini, Perplexity, and Grocke AI, which is the best
Anthropic also announced the launch of a cloud code SDK, which allows users to manufacture their own AI-operated tools and agents, while taking advantage of the same “core agent” as a cloud code to ensure that they achieve the same level of assistance. As an example, Anthropic shared the launch of the cloud code on Github in Beta, which allows users to call on the cloud code on PRS (bridge requests) to assist errors, responds to the reviewer response, and more.
Get top stories of morning with us in your inbox every day Tech Today Newsletter.
