During its inaugural developer conference on Thursday, Anthropic launched two new AI models that the claims of the startup are one of the best in the industry, at least how they score on the popular benchmark.
According to the company, Cloud Opus 4 and Cloud Sonnet 4, new family of anthropic, part of Cloud 4, can analyze big data sets, execute long-delaying tasks, and can take complex action. Both models were designed to perform well on programming tasks, saying anthropic, they were well suited for writing and editing codes.
Both users and users of the company's free chatbot apps will get access to Sonnet 4, but only the payment users will get only OPUS 4 access. For API of anthropic, through Amazon's bedock platform and Google's Wartax AI, OPUS 4 will be priced at $ 15/$ 75 per million tokens (input/output) and will be more than $ 3/$ 15 on Sonnet 4.
There are raw bits of token data that work with the AI model, with a million tokens equal to about 750,000 words – “war and peace” for a long time about 163,000 words.
Anthropic Cloud 4 models arrive as the company significantly increases revenue. Rentally, the outfits established by pre-Opnea researchers aim to earn $ 12 billion in earnings in 2027, estimated at $ 2.2 billion this year. Anthropic recently discontinued a $ 2.5 billion credit facility and raised billions of dollars from Amazon and other investors in anticipation of increasing costs associated with the development of Frontier models.
The rivals have not made it easy to maintain the position of the pole in the AI race. While Anthropic launched a new flagship AI model earlier this year, with Cloud Sonnet 3.7, with an agent coding tool, called Cloud Code, contestants, including OpenI and Google, have raced to carry forward the company with powerful models and their own devas tooling.
Playing with anthropic cloud 4.
Today, which are more capable of two models, Opice 4, can maintain “concentrated efforts” in several stages in a workflow, call anthropic. Meanwhile, the sonnet is designed as a “drop-in replacement” for 4-sannette 3.7-coding and mathematics improves compared to previous models of angroopic and follow the instructions more accurately.
Cloud 4 Family is also less likely to attach to “reward hacking”, less than 37, anthropic claims. Reward hacking, also known as specification gaming, is a behavior where models take shortcuts and flaws to complete tasks.
To be clear, these reforms have not yielded the world Best Model by every benchmark. For example, while Ops 4 has verified the Swe-4 and GPT-4.1 of the Gemini 2.5 Pro of Google and the GPT-4.1, designed to evaluate the coding capabilities of a model, this multimodal assessment cannot cross the O3 on MMMU or GPQA Diamond, PHD-Level Biology- A set of physics-chemistry science.
Nevertheless, the anthropic strict security measures are issuing Ops 4, including beef-up harmful material detector and cyber security defense. The company claims that its internal testing found that Opus 4 can “greatly increase the capacity of a person with a stem background to get chemical, biological or nuclear weapons to receive or deploy,” ASL -3 “model of anthropic reaches the model specification.
Both Opus 4 and Sonnet 4 are “hybrid” models, anthropic says an extended thinking for the enabled and intense logic (to the extent AI “causes” and “thinking” because humans understand these concepts). With the logic switching to the logic mode, the model may take longer to consider the possible solutions to a problem before responding.
As due to the model, they will show a “user-friendly” summary of their thought process, anthropic. Why didn't you show the whole thing? To protect the “competitive advantage” of the anthropic, partially, the company accepts in a draft blog post provided to Techcrunch.
Opus 4 and sonnet 4 can use many devices, such as the search engine, in parallel, and optional between arguments and equipment to improve the quality of their answers. They can remove and save the facts in “memory” to handle the tasks more strongly, which over time create anthropic described as “silent knowledge”.
To make the model more programmer-friendly, the anthropic is rolling out the upgrade to the aforesaid cloud code. Cloud code, which allows developers to run specific functions directly through the model of anthropic from the terminal, now integrates with IDES and provides an SDK that allows the gods to connect it with third-party applications.
Cloud Code SDK announced earlier this week, enables to run cloud code as a sub-process on the supported operating system, which provides a way of manufacturing AI-operated coding assistants and equipment that take advantage of the capabilities of the cloud models.
Anthropic has issued cloud code extensions and connectors for Microsoft's VS code, jetbrains and github. The Github connector allows developers to tag the cloud code to respond to the reviewing response, as well as try to fix the errors – or otherwise modified – code.
The AI models still struggle for code quality software. Code-generating AIs, security weakens and errors, due to weaknesses in areas such as the ability to understand programming logic. Nevertheless, his promise to promote coding productivity is pushing companies to adopt companies and developers.
The anthropic, promising intensely aware of it, is promising more frequent models updates.
“Were […] Stir for more frequent models updates, providing a stable stream of reforms that bring customers the ability to success rapidly, “Wrote a startup in your draft post.” This approach keeps you on the cutting edge because we constantly refine and enhance our model. ,