OpenAI Unveils GPT-4.1, Its Flagship AI Model

OpenAI has released GPT-4.1, an updated version to the GPT-4o multimodal AI model released by the firm last year. During a webcast on Monday, OpenAI stated GPT-4.1 has an even broader context window and is superior to GPT-4o in “just about every dimension,” with huge gains to coding and instruction following. OpenAI further unveiled the GPT-4.1 family of models. Indeed, “4.1” as if the company’s names weren’t already confused enough.

Developers may now access GPT-4.1 and two smaller model variants. These include the even more lightweight GPT-4.1 Nano, which OpenAI claims is its “smallest, fastest, and cheapest” model to yet, and the GPT-4.1 Mini, which is, like its predecessor, more reasonably priced for developers to experiment with.

OpenAI claims that the GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano “excel” at coding and obeying instructions. The multimodal models, which are accessible via OpenAI’s API but not ChatGPT, can process around 750,000 words in a single session, which is longer than “War and Peace.” They have a 1-million-token context window.

Close to one million tokens of context, such as the text, pictures, or videos that are part of a prompt, can be processed by any one of the three models. That is a lot more than the 128,000 token cap on GPT-4o. In an article revealing the models, OpenAI states, “We trained GPT‑4.1 to reliably react to information over the full 1 million context length.” Additionally, it has been taught to be significantly more accurate than GPT-4o in identifying pertinent content and ignoring distractions for both short and long context durations.

As OpenAI competitors like Google and Anthropic step up their attempts to create complex programming models, GPT-4.1 is released. On well-known coding benchmarks, Google’s recently released Gemini 2.5 Pro, which also features a context window with a million tokens, scores highly. So do the updated V3 from DeepSeek, a Chinese AI firm, and Claude 3.7 Sonnet from Anthropic.

Additionally, GPT 4.1 is 26% less expensive than GPT-4o, a parameter that has gained significance since DeepSeek’s incredibly effective AI model was introduced.

The debut coincides with OpenAI’s announcement in a changelog that GPT-4o is a “natural successor” to its two-year-old GPT-4 model, which it intends to phase out from ChatGPT on April 30th. Given that “GPT-4.1 offers better or comparable performance on many essential features at much lower cost and latency,” OpenAI also intends to deprecate the GPT-4.5 beta in the API on July 14.

In order to avoid its GPUs from “melting,” OpenAI had to restrict requests and suspend access to free ChatGPT accounts after updating GPT-4o, the default model in ChatGPT, last month to provide additional image-generation capabilities.

The GPT-4.1 announcement represents a change in OpenAI’s release timetable and validates our information from last week that the business is getting ready to introduce new models. CEO Sam Altman said on X on April 4th that the GPT-5 launch was being postponed and will now take place “in a few months,” which is later than the originally anticipated May target. According to Altman, OpenAI “found it more difficult than we anticipated it would turn out to be to seamlessly incorporate everything,” which is one reason for the delay.

The complete version of OpenAI’s o3 reasoning model and an o4 micro reasoning model are expected to be released soon; AI engineer Tibor Blaho has already seen references to them in the most recent ChatGPT online release.