Openai today announced that it is releasing a new family of artificial intelligence models optimized for superior coding to enhance efforts to dodge increasingly tough competition with companies like Google and Anthropic. This model is available to developers through OpenAI’s Application Programming Interface (API).
Openai has released three size models: the GPT 4.1, GPT 4.1 Mini, and GPT 4.1 Nano. Openai’s Chief Product Officer Kevin Weil said in the live stream that the new model is better than Openai’s most widely used model, the GPT-4O, and in some respects it’s biggest and most powerful model, GPT-4.5.
GPT-4.1 won 55% on the SWE Bench, a benchmark widely used to measure the prowess of coding models. The score is a few percentage points over the scores of other Openai models. The new model is “excellent in coding and excellent in complex teaching.
The ability of AI models to write and edit code has improved significantly over the last few months, making software prototyping more automated, and the capabilities of so-called AI agents have improved. Rivals like Anthropic and Google have introduced models that are particularly suitable for writing code.
The arrival of the GPT-4.1 has been widely rumored for weeks. Openai appears to have tested the model on several popular leaderboards under the pseudonym Alpha Quasar, sources say. Some users of the “stealth” model It has been reported Impressive coding ability. “Quasar fixed all open issues where other code was adopted. [sic] Through an incomplete LLMS,” one person wrote on Reddit.
All new models can analyze 8x more code at a time, making improvements and improving your ability to fix bugs. The new model also makes it better to follow user-directed instructions, reducing the need to repeat commands in various ways to achieve the desired result. Openai demonstrated a GPT-4.1 demo that builds a variety of apps, including flashcard apps for language learning.
“Developers are very interested in coding. They have improved the ability to write functional code,” Michelle Pokras said in a live stream on Monday after training at Openai. “We’ve been working to follow a variety of formats, explore repos better, run unit tests, and write code to compile.”
GPT-4.1 is 40% faster than GPT.4o, the most widely used by developers, which is the most widely used by GPT.4o. This latest version reduces the cost of entering user input queries by 80%, says Openai.
In today’s livestream, Varun Mohan, CEO of Windsurf, a popular AI coding tool, found that the company is testing the GPT-4.1 and that the new model is “60% better” than the GPT-4o, according to its own benchmark. “We found that there were significantly fewer cases of degenerate behavior in GPT-4.1,” Mohan said, noting that the new model would incorrectly spend less time reading and editing unrelated files.
Over the past few years, Openai has attracted enthusiasm for ChatGpt, the incredible chatbot first announced in late 2022, selling a business selling access to more advanced chatbots and AI models. In a TED interview last week, Altman said Openai has 5 million active users per week and its usage is “growing very rapidly.”
