OpenAI Launches New GPT-4.1 Models, But They’re Not Coming to ChatGPT

OpenAI has launched a new series of AI models including GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. They are the best non-reasoning AI models from OpenAI, but the GPT-4.1 series is not coming to ChatGPT. OpenAI says GPT-4.1 series models are exclusively trained for developers, and they are available in the API, starting today. GPT-4.1 was tested on OpenRouter as “Quasar Alpha”.

During the launch, OpenAI CPO Kevin Weil said GPT-4.1 is “better than GPT-4o on just about every dimension.” And on some key benchmarks, the GPT-4.1 model matches or beats the massive GPT-4.5 AI model. OpenAI says GPT-4.1 excels at coding, instruction following, and long context retrieval.

The new GPT-4.1 AI models come with a context window of 1 million tokens for the first time, which is huge, and have a recent knowledge cutoff of June 2024. So in many ways, the GPT-4.1 model looks like a major upgrade to the GPT-4o model. Not to mention, these are multimodal models which means you can process both text and images as well.

Coming to benchmarks, OpenAI says on SWE-bench Verified, GPT-4.1 completed 55% of tasks, much higher than GPT-4o’s 33% and o3-mini-high’s 49%. Next, in instruction following, GPT-4.1 came very close to GPT-4.5, o3-mini-high, and o1-high on OpenAI’s internal benchmark.

Following that, on the Needle in a Haystack test, which evaluates information retrieval in a long context window, all three models under the GPT-4.1 series correctly retrieved the information up to 1 million tokens.

Finally, coming to pricing, GPT-4.1 costs $2/$8 for input/output tokens per 1 million tokens. Basically, in median queries, GPT-4.1 is 26% cheaper than GPT-4o while offering much better performance. GPT-4.1 nano is the cheapest and fastest model by OpenAI, ever. Lastly, OpenAI mentioned that GPT-4.5 is going away from the API on 14th July, 2025.