Google returned to the path of victory in the AI models with Gemini 2.5 Pro, but has been surpassed for a couple of days with Claude Opus 4 and Claude Sonnet 4the two new Anthropic models that raise the bar with new goals achieved.
Claude Opus 4 becomes the best coding model thanks to its huge Ability to manage complex tasks and the workflows of agents AI. While Claude Sonnet 4 is a significant update to Claude Sonnet 3.7, the latest model that was released in February.
In Anthropic they have reasons for joy, not only for these two models, but because now it includes the possibility of Use “extended thought” and even search the web at the same time. Claude can now use one or the other to give better answers.
Claude Opus 4 and sonnet 4 are hybrid models which offer both instant responses and extended thinking for more complex reasoning. And the one who stays out of being free is the first, so Sonnet 4 is already available in Spain, although in a limited way as Sonnet has been according to Tokens that are used.
Claude Sonnet 4 improves sonnet 3.7 in coding with 72.7 % in Swe-Bench. The key to the new model is in its ability to balance performance and efficiency for external and internal cases. It does not reach Opus 4’s ability in many of its domains, but offers great experience in general terms.
The performance of both Anthropic models
Claude Opus 4 becomes the highest performance model until the Anthropic date and the best programming model By exceeding Gemini 2.5 Prowhich surprised his own and strangers with the big step taken by Google since he deployed it. Opus 4 figures are 72.5 % in Swe-Bench and 43.2 % terminal.
One of its best skills is in the performance of tasks that take a long time and that require continued effort and thousands of steps, and with the extra that means that it works continuously for many hours. In coding it becomes the number one model current, especially in the understanding of complex code bases.
There are more, and Anthropic has also announced the new capabilities of both models to use tools in parallel, follow instructions more precisely and give access to local archives for developers. In fact, Claude Opus 4 will only be available through the subscription PRO, Max, Team and Enterprise in Spain.
There are two interesting details in the improvements of the two models: the behavior of the use of direct access or Loopholes. Opus 4 ha drastically improved memory capacity by being able to keep the memory files that he accessed locally on the PC.
Results of both models in software engineering
In summary, and as Anthropic maintains from his ad on its website, Sonnet 4 brings better performance for daily tasks As an important update from Sonnet 3.7, and Opus 4 exceeds the limits in coding, research, writing and scientific discovery.
In this way, the wheel in which Openai, Google and Anthropic are overcoming each other Every time they launch a greater update to their AI models. Now the play of the creators of Chatgpt is close to the GPT-5 announcement.
