New GPT 4.1 models from Openai excelled in coding

Openai Today she announced that it is issuing a new family from artificial intelligence Models for excellence in coding have been improved, intensifying the efforts made to increasingly rivalry in companies such as Google and Anthropic. Models are available for developers through the Openai (API) application interface.

Openai releases three sizes of models: GPT 4.1, GPT 4.1 Mini and GPT 4.1 Nano. Kevin Whale, chief product employee at Openai, said on one end, the new models are better than the most used model in Openai, GPT-4O, and the best and strongest model, GPT-4.5, in some respects.

GPT-4.1 55 percent recorded on Swe-Pect, a widely used standard to measure the ingenuity of coding models. The result is several percentage points higher than other Openai models. Will said the new models “are great in coding, they are great in following complex instructions, it’s great for construction agents.”

The ability of artificial intelligence models to write and edit the code dramatically in recent months, allowing more automatic ways for primary models programs and improving the capabilities of the so -called Artificial intelligence agents. Comledrants like man and Google Provide both models that are particularly good in writing code.

GPT-4.1 has widely common for weeks. Openai appears to have tested the model on some popular top boards under the ALPHA Quasar pseudonym, the sources say. Some users of the “hide” model I mentioned Impressive coding capabilities. One person wrote to Redait: “Quasar has fixed all the open problems she faced with another SIC code (SIC) via LLMS that was incomplete,” said one of the people.

All new models can analyze the code eight times at a time, which improves their ability to make improvements and repair errors. New models are also better in the following guidelines provided by users, which reduces the need to repeat orders in different ways to get the desired result. Openai showed experimental offers for GPT-4.1, building different applications including the Flashcard language card for language learning.

“The developers are very interested in coding, and we have improved our model’s ability to write functional programming instructions,” said Michelle Boukras, who operates in the post -training period in Openai, during Livestream on Monday. “We have made it tracking different formats, exploring restarts, operating unit tests, and writing code that collects.”

GPT-4.1 is 40 percent faster than GPT.4O, the most used Openai model for developers. Openai says that the cost of users who enter queries has been reduced by 80 percent in this latest version.

On the live broadcast today, Varun Mohan, CEO of Windsurf, a common tool for the coding of artificial intelligence, said that the company was testing GPT-4.1 and found that the new model “60 %” is better than GPT-4O according to its criteria. “We have found that GPT-4.1 has much lower cases of deteriorating behavior,” Mohan said, noting that the new model spends less time reading and editing unrealistic files by mistake.

Over the past two years, Openai has made a feverish attention in ChatgptChatbot is great The veil was revealed for the first time in late 2022In increased business sells access to more advanced chat and artificial intelligence models. In an interview with TED last week, Altman said that Openai had 500 million active weekly users, and that the use “grows very quickly”.