OpenAI announced the new GPT-4.1 model along with two smaller versions during a livestream on Monday, one year after the launch of its preceding GPT-4o model. The company said that GPT 4.1 has a bigger context window, claiming that it is an improvement over GPT-4o in “just about every dimension.”
The new 4.1 model is available for developers alongwith the smaller models, GPT-4.1 Mini and GPT-4.1 Nano. The mini AI model is a more affordable option for developers to modify, while the nano version is a lightweight model OpenAI says is its “smallest, fastest, and cheapest.”
In a post announcing the new model, OpenAI said, “We trained GPT‑4.1 to reliably attend to information across the full 1 million context length.” All three models can process up to one million tokens of context, including text, images and videos in a prompt. This is a major improvement over GPT-4o’s 128,000 token limit. Compared to its predecessor, the model has also been trained to pick up more relevant text and ignore distracting factors in long and short context prompts.
GPT-4.1 is priced 26 per cent cheaper than GPT-4o and the launch arrives as the AI company aims to phase out the two-year-old GPT-4. The news was announced in a changelog stating that upgrades to GPT-4o make it a natural replacement. OpenAI plans to discontinue the model due to, “improved or similar performance on many key capabilities at much lower cost and latency,” of the GPT-4.1 model.
ChatGPT-4o, the current default version, received an update last month and added new image generation capabilities to the AI model. The new feature gained popularity recently, forcing OpenAI to limit image generation requests and put free accounts on hold to prevent their GPUs from melting.
The CEO of OpenAI, Sam Altman, had also announced on April 4 that the launch of GPT-5 had been pushed back and was to arrive “in a few months.” Altman said that the delay is partially because the company “found it harder than we thought it was going to be to integrate everything smoothly.” OpenAI is set to release the full version of its o3 and o4-mini reasoning model in a, “couple of weeks,” while AI engineer Tibor Blaho found references in its latest web release.
In the News: Cybercrime group rolls out major updates for on sale phishing tool
