OpenAI has unveiled its latest flagship AI model, GPT-4.1, an upgraded version of GPT-4o. The new model shows significant improvements across multiple dimensions, particularly in coding capabilities and instruction-following accuracy.

Alongside GPT-4.1, OpenAI has also introduced two lightweight models: GPT-4.1 Mini and GPT-4.1 Nano. The Mini version maintains the cost-effectiveness of its predecessor, making it ideal for developer experimentation and application development. Meanwhile, the Nano variant is described by OpenAI as its "smallest, fastest, and most affordable" model to date.

All three models boast the ability to handle up to one million tokens in context length, a substantial increase from GPT-4o's limit of 128,000 tokens. According to OpenAI, GPT-4.1 has been trained to reliably process information of this magnitude, demonstrating enhanced reliability in identifying relevant text and filtering out irrelevant data.

In terms of pricing, GPT-4.1 offers a 26% reduction compared to GPT-4o, a move that aligns with recent advancements in efficiency brought about by competing models like DeepSeek's offerings.

With this launch, OpenAI plans to phase out the GPT-4 model, which was released two years ago, discontinuing its use in ChatGPT starting April 30. Additionally, the company will stop providing the GPT-4.5 preview via API on July 14, as GPT-4.1 delivers improved or comparable performance at a lower cost and with reduced latency.

Notably, GPT-4o, currently the default model for ChatGPT, received an update last month that introduced advanced image generation features. These enhancements proved so popular that OpenAI had to impose request limits and temporarily suspend access for free-tier ChatGPT accounts to prevent GPU overload.

The release of GPT-4.1 confirms recent reports about OpenAI’s preparation for a new model rollout and signals a shift in the company’s release strategy. This comes after OpenAI CEO Sam Altman announced that the launch of GPT-5 would be delayed until "a few months later," pushing back the previously anticipated May deadline. Altman cited challenges in integrating all components seamlessly as a key reason for the delay.

Looking ahead, OpenAI is also preparing to roll out the full version of its o3 inference model and the o4 mini inference model, as hinted at in recent updates discovered by AI engineer Tibor Blaho within the latest ChatGPT web interface.