OpenAI has officially unveiled GPT-4.1, a new iteration of the GPT-4o multimodal AI model introduced last year. In a livestream held on Monday, the company announced that GPT-4.1 offers a significantly enlarged context window and outperforms GPT-4o in nearly every aspect, particularly excelling in coding capabilities and adherence to user instructions.
Developers now have access to GPT-4.1, alongside two more compact versions: GPT-4.1 Mini, designed to be cost-effective for experimentation, and GPT-4.1 Nano, touted as OpenAI’s “smallest, fastest, and cheapest” model available to date.
All three models possess the capability to handle contexts of up to one million tokens, a substantial increase compared to GPT-4o’s 128,000-token limit. “We trained GPT-4.1 to consistently attend to pertinent information throughout the entire one million token context,” OpenAI highlighted in a recent announcement. “Additionally, it has been trained to be significantly more adept than GPT-4o at identifying relevant text while disregarding distractions, regardless of context length.”
Furthermore, GPT-4.1 is 26% more cost-effective than GPT-4o, a consideration that has gained importance following the introduction of DeepSeek’s highly efficient AI model.
This release coincides with OpenAI’s plan to retire the two-year-old GPT-4 model on April 30, as noted in a changelog that describes the recent enhancements to GPT-4o as a “natural successor” for GPT-4. Additionally, OpenAI intends to phase out the GPT-4.5 preview in its API on July 14, citing that GPT-4.1 provides enhanced or comparable performance at a much more competitive price and reduced latency.
The current default model in ChatGPT, GPT-4o, received an update last month that included new image-generation features, which proved so popular that OpenAI had to impose restrictions on requests from free accounts to safeguard its GPU infrastructure.
The launch of GPT-4.1 aligns with earlier reports regarding OpenAI’s preparations for new model deployments and indicates a shift in the company’s product schedule. On April 4, CEO Sam Altman announced via X that the launch of GPT-5 would be delayed until “in a few months,” later than the initially anticipated May timeline. According to Altman, this setback is partly due to challenges in ensuring seamless integration.
OpenAI also aims to soon release the full version of its o3 reasoning model along with an o4 mini reasoning model, with references to these models already detected in the latest ChatGPT web application update.
