Cute, which is a vibe coding tool, says Cloud 4 has reduced its errors by 25% and increased it rapidly by 40%.
On 22 May, Anthropic began to roll two new models: Cloud Sonnet 4 and Cloud Ops 4. While the sonnet is available for free users, opus requires a payment subscription and is able to improve the sonnet when it comes to coding.
One in blog postAnthropic confirmed that Cloud Oppus 4 scored 72.5 percent in Swe-Bench (SWE is small for software engineering benchmark).
In trials, Opace 4 gave a continuous performance on long -running works, requiring concentrated efforts and thousands of stages.
Anthropic also claimed that its newest model worked directly on the code for seven hours.
Vibe coding company Luvobal, which uses clouds in its “AI-AI-Integrated Prompt-based web and apps builder” tools, has seen similar reforms after upgrading Cloud 4.
One in Post On X, Luvable says that it has 25% less errors and all projects (including old projects) are 40% faster after the total projects and Edit are 40% faster after deploying Cloud 4.
One in Different postsCute founder Anton Osika confirmed that “Claude 4 erased the most cute errors” while especially mentioned the LLM syntax errors during vibi coding.
Cloud 4 is a good model for coding
While the opinion on Claude 4 is mixed, I have personally seen that Claude 4 produces code with less errors than Gemini when I am working on the Dart/Kotlin app.
It also depends on the project and reference to the project, but in projects where a long reference is not required, Claude 4 improved Gemini in my tests.
The cloud model has always maintained the reputation of “best at coding”, but has recently been very competing with Google, which released Gemini 2.5 Pro with 1 million reference window.
Compared to Cloud 4 or 200,000 reference windows of old models, 1 million reference window for Gemini 2.5 gives it an advantage. But it is not necessary that Gemini 2.5 is better than Cloud 4 in coding.
Both can be surprisingly luxurious and even terrible at the same time, and it also comes down how you do the signal engineering.
Mixing models is always good, such as O3 or Gemini for planning and cloud 4 and Gemini for coding.
Based on the analysis of 14M malicious tasks, search for the top 10 MITERAT & CK techniques behind the 93% attacks and how to defend them against them.
