OpenAI Releases GPT-4.1 Model

On April 15, OpenAI introduced GPT-4.1, the next version of the multimodal AI model GPT-4o that was released last year. During a livestream introducing the new model, OpenAI said that GPT-4.1 has a larger context window and outperforms GPT-4o “in nearly every aspect,” especially with major improvements in programming and following instructions.

OpenAI launches GPT-4.1 model, powerful in programming. Illustration photo: Getty Images

GPT-4.1 is now available to developers, along with two smaller versions, GPT-4.1 Mini and GPT-4.1 Nano. The Mini is a lower-cost version that makes it easier for developers to experiment, while the Nano is even lighter, and Open AI claims it is the “smallest, fastest, and cheapest” ever.

All three models can handle up to 1 million context tokens – including text, images, or videos – far exceeding the 128,000 tokens limit of GPT-4o. “We trained GPT-4.1 to consistently retrieve scene information across the entire context length of 1 million tokens,” Open AI said in a post introducing the new AI models. “We also trained it to be much more reliable than GPT-4o at recognizing relevant text and ignoring noise,” the company added.

GPT 4.1 is also 26% cheaper than GPT-4o, a number that matters a lot after DeepSeek launched its ultra-efficient AI model.

The launch comes as OpenAI prepares to remove the GPT-4 model from ChatGPT at the end of April. The company claims that the improvements to GPT-4o make it a “natural successor” to GPT-4, which was released two years ago. With the release of GPT-4.1, OpenAI also plans to discontinue support for the developer API preview of GPT-4.5 on July 14. “GPT-4.1 delivers better or equivalent performance on many important features, with lower cost and latency,” OpenAI says.

How to Win Friends

(translated and compiled from The Verge)

Source: https://baodongnai.com.vn/xa-hoi/khoa-hoc-cong-nghe/202504/openai-ra-mat-mo-hinh-gpt-41-6be1424/