• Source:JND
Add as a preferred

OpenAI released their latest series of AI models, which includes GPT-4.1, GPT-4.1 Mini, and GPT-4.1 Nano. The models are said to perform better in coding tasks and understanding long prompts. They also added up to 1 million tokens in context, better following instructions, and improved accuracy for the advanced AI applications being developed. However, these models have not yet been integrated into ChatGPT. Instead, they are being steadily integrated into the OpenAI API. ChatGPT users will not notice immediate improvements, but OpenAI has stated that multiple GPT-4.1 capabilities will be incorporated into future versions of GPT 4.0, meaning upgrades will be rolled out after some time.

GPT 4.1

Under the name Quasar Alpha, GPT-4.1 was integrated into OpenRouter. The last non-reasoning model of OpenAI, the GPT-4.1 series, is restricted for API users only.

While speaking at the launch, OpenAI’s CPO Kevin Weil highlighted that GPT-4.1 is “better than GPT-4o on just about every dimension.” He noted that on some key benchmarks, the GPT-4.1 model matches or surpasses the capabilities of the colossal GPT-4.5 AI model. Performatively, OpenAI claims that GPT-4.1 shines in coding, following instructions, and retrieving information from lengthy contexts.

For the first time, the new GPT-4.1 AI models have a context window of 1 million tokens, which is unprecedented, along with a recent knowledge cutoff of June 2024. In many aspects, the GPT-4.1 model seems quite advanced in comparison to the GPT-4o model. Additionally, these are multimodal models, meaning that text and images can be processed simultaneously.  

Regarding benchmarks, OpenAI states that in SWE-bench Verified, GPT-4.1 completed 55 percent of tasks, significantly outperforming GPT-4o’s 33 percent and o3-mini-high’s 49 percent. Then, in instruction following, GPT-4.1 managed to get extremely close to GPT–4.5, o3-mini-high, and o1-high in OpenAI’s internal benchmark.  

In the Needle in a Haystack test, which evaluates information retrieval within a lengthy context window, all three models belonging to the GPT-4.1 series managed to retrieve information correctly up to 1 million tokens.

Lastly, regarding pricing, GPT-4.1 is charged $2/$8 for input and output, respectively, per 1 million tokens. In median queries, GPT-4.1 is 26% cheaper than GPT-4o, with much better performance. GPT-4.1 nano is the model deemed the cheapest and fastest by OpenAI, ever. Lastly, OpenAI said GPT-4.5 is going away from the API on July 14, 2025.

That was it, guys, for this one, keep an eye out on English Jagran for more such updates!