The latest versions of anthropic Claude Generative models of AI They made their debut on Thursday, including a heavier model built specifically for coding and complex tasks.
Anthropic launched the new Claude 4 opus and Claude 4 sonnet During his code, models with Claude's and directors conference said the new tools mark a significant step forward in terms of reasoning and deep thinking skills.
The company launched the previous model, Claude 3.7 sonnet, in February. Since then, AI competitors have also stepped up their game. Open Released GPT-4.1 in AprilWith an emphasis on an extended context window, along with the new O3 model of judgment family. Google followed in early May with Updated version of Gemini 2.5 Pro That he said it was better in coding.
Claude 4 opus is a larger, more intense resource model, built to deal with particularly difficult challenges. Anthrop chief executive Dario Amodei said the test users saw that he quickly tackled the tasks that a person might need to complete them for several hours.
“In many ways, as we often find in large models, the benchmarks do not make complete justice for it,” he said during Introductory event.
Anthropic said Claude 4 noticed how to move while playing Pokémon.
Sonnet Claude 4 is a lean model, with improvements built on the model of Anthropian Claude 3.7. Model 3.7 often had problems with excessive and sometimes did more than the person asked to do, Amodi said. Although it is less model intensive with resources, it is still going well, he said.
“It actually does the same as an opus of some of the coding benchmarks, but I think it's lean and narrower,” Amodi said.
Anthropic said models have a new opportunity, are still being tested in beta stage, in which they can use tools such as web searches while involved in prolonged reasoning. Models can alternate between judgment and tools to get better answers to complex searches.
Models offer almost non -permanent response modes and prolonged thinking modes.
All paid plans offer Opus and Sonnet models, while the free plan only has the Sonnet model.
The new models show the focus of Anthropic on building strong coding models, said Arun Chandrasekaran, a prominent vice president, analyst at Gartner. “The Claude Anthropic models have established strong leadership in the domain of software engineering and the latest edition of Claude 4 expands that leadership.”
Anthropic activates security protocols with new Claude models
At the launch of the Cloude Opus 4 model, Anthropic said he was taking increased security measures to reduce the risk of Claude's abuse. In a Blog -PostThe company said it did not determine whether the model actually requires the protection of its ASL-3 standard, but it does so as caution.
Safety precautions are specially designed to prevent Claude to help develop chemical, biological, radiological or nuclear weapons. Anthropic said it restricts attacks known as universal prisoners that allow attackers to overcome existing protocols. “We will continue to assess the capabilities of CBRN's Clude Opus 4,” the Anthropic blog said. “If we conclude that Claude Opus 4 did not exceed the relevant threshold of ability, then we can remove or adjust the ASL-3 protects.”
Chandrasekaran said the implementation of security standards is worth noting. “This includes improved computer security measures and fast classifiers to mitigate the risks associated with powerful AI systems,” he said. The new models show the company's focus on balancing new security technology, he said.