Openai has announced today that he launches a new family of optimized artificial intelligence models to codify, as he increases efforts to defend the increasingly rigid competition of companies like Google and Anthropic. Models are available to developers through the Openai App (API) application program.
Openai throws three models sizes: GPT 4.1, GPT 4.1 Mini and GPT 4.1 Nano. Kevin Weil, Openai Product Director, said in a live streaming that the new models are better than the most commonly used model of Openai, GPT-4o, and better than their largest and most powerful model, GPP-4.5.
GPT-4.1 scored 55 percent in Swe-Cench, a widely used point of reference for the ability to encode coding models. The score is from several percentage points above that of other Openai models. The new models are “excellent in coding, they are excellent in complex instructions, they are fantastic for agents,” said Weil.
IA’s models for writing and editing the code has improved significantly in recent months, allowing more automated ways to prototype software and improve the skills of the so -called AI agents. Rivals like Anthropic and Google have entered models that are especially good for writing code.
The arrival of GPT-4.1 has been widely rumored for weeks. Openai apparently tested the model in some popular classification tables under the pseudonym Alpha Quasar, according to sources. Some users of the “furtive” model reported impressive coding skills. “Quasar solved all the open problems he had with other Code generate (sic) through Llms that was not complete,” a person wrote in Reddit.
All new models can analyze eight times more code at the same time, which improves their ability to improve and solve errors. New models are also better in the instructions that users give, reducing the need to repeat orders in different ways to obtain the desired result. Openai showed demonstrations of GPT-4.1 Creation of different applications, including a flashcard application for language learning.
“Developers are very concerned about coding and we have improved the ability of our functional writing model,” said Michelle Pokrass, who works in post-training training in Openai during Monday’s streaming. “We have been working to continue following different formats and explore better reposts, run units and write code that is compiled.”
GPT-4.1 is 40 percent faster than GPT.4O, the most used model for Openai for developers. The cost of users who enter queries has been reduced by 80 percent in this latest version, according to Openai.
At the streaming Livesream, Varun Mohan, CEO of Windsurf, a popular tool for the CODIFICATION of the AI, said that the company had been trying GPP-4.1 and found that the new model was “60 percent” better than GPT-4O according to its own reference points. “We have found that GPT-4.1 has substantially fewer cases of degenerated behavior,” said Mohan, emphasizing that the new model spends less time reading and editing irrelevant files by mistake.
Over the last two years, Openai has spoken feverish interest in Chatgpt, a notable chat that was first presented at the end of 2022, in a growing business that sold access to more advanced chat models and AI models. In an interview TED last week, Altman said that Openai had 500 million active weekly users and that the use was “growing very quickly”.
