(Reuters) – OpenAI on Monday launched its new AI model GPT-4.1, along with smaller versions GPT-4.1 mini and GPT-4.1 nano, touting major improvements in coding, instruction following, and long context comprehension.

The new models, available only on OpenAI’s application programming interface (API), outperform the company’s most advanced GPT-4o model across the board, the ChatGPT maker said.

With improved context understanding, they can support up to 1 million “tokens” — a term that refers to the units of data processed by an AI model. The models are also equipped with refreshed knowledge up to June 2024.

GPT-4.1 showed a 21% improvement over GPT-4o and 27% over GPT-4.5 on coding. Meanwhile, the improvements in instruction following and long context comprehension also make the GPT-4.1 models more effective at powering AI agents.

“Benchmarks are strong, but we focused on real-world utility, and developers seem very happy,” CEO Sam Altman said in a post on social media platform X.

The family of models also operate at a “much lower cost” compared to GPT-4.5, OpenAI said. The company added it would turn off the GPT-4.5 preview that is available in the API in July, as the new models offer “improved or similar performance.”

OpenAI in February released the GPT-4.5 research preview for some users and developers and announced plans to expand access in subsequent weeks.

(Reporting by Deborah Sophia in Bengaluru; Editing by Alan Barona)