Site iconSite icon ForkLog

OpenAI Unveils Programming-Focused GPT-4.1

OpenAI Unveils Programming-Focused GPT-4.1

OpenAI has introduced a new family of AI models: GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. These models “excel” at programming and following instructions.

The neural networks are available via API. They feature multimodal capabilities and have a context window of a million tokens (capable of processing approximately 750,000 words at once).

This new release marks a step towards creating AI models capable of developing entire applications, ensuring high quality, error testing, and documentation writing.

GPT-4.1 surpasses GPT-4o and GPT-4o mini in benchmarks evaluating programming, including SWE-bench. GPT-4.1 mini and nano are more efficient and faster due to reduced precision.

OpenAI highlighted that GPT-4.1 nano is the fastest and cheapest model in history. The pricing is as follows:

According to the company’s internal tests, GPT-4.1 scored between 52% and 54.6% in SWE-bench Verified and 72% in the Video-MME context understanding category for “long video without subtitles.”

Comparison of GPT-4.1 with other models in the SWE-bench Verified test, assessing the ability to solve real-world software development tasks. Data: OpenAI.

OpenAI announced that on July 14, it will discontinue access to GPT-4.5 — its largest AI model — via API. Developers need to switch to another company solution. GPT-4.1 is positioned as the preferred replacement.

“[GPT-4.1] offers similar or improved performance compared to GPT-4.5 in key areas at a much lower cost,” the startup emphasized.

OpenAI charges $75 per million tokens input into GPT-4.5, and $150 per million tokens generated by it. This is 30 times the cost of input data and 15 times the cost of output data compared to GPT-4o.

GPT-4.5 will remain in ChatGPT for users with paid subscriptions.

In April, GPT-4.5 passed a tripartite Turing test.

Exit mobile version