OpenAI launches three new GPT-4.1 models via its API, which are said to excel in the area of coding. The largest AI model to date, GPT-4.5, has to make way and is being phased out in the API.
OpenAI has made three new GPT-4.1 models available through its API. The models perform better than their predecessors in terms of coding, instruction following, and processing long contexts. With GPT-4.1, developers gain access to higher performance at lower costs. At the same time, the company announces the phasing out of its largest AI model GPT-4.5 from the API. GPT-4.1 thus presents itself as a worthy replacement.
Improved performance
The new models are GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. All three support context windows of up to 1 million tokens. This means they can process larger amounts of text and code, which is interesting for legal, technical, and financial applications, among others. According to OpenAI, GPT-4.1 shows better performance on benchmarks than GPT-4o and GPT-4.5.
On the SWE-benchmark, which measures how well a model performs programming tasks, GPT-4.1 achieves a score of 54.6 percent. That’s 21.4 percentage points higher than GPT-4o. GPT-4.1 also proves to be more reliable in following instructions. The model scored 38.3 percent on the MultiChallenge benchmark, an increase of 10.5 percentage points compared to GPT-4o.
GPT-4.1 mini offers similar performance to GPT-4o, but at lower costs and with lower latency. GPT-4.1 nano focuses on speed and efficiency. The model is suitable for lighter tasks such as text classification or auto-completion, and is the cheapest of the three.
Applications
According to OpenAI, the GPT-4.1 models are optimized based on feedback from developers. In practice, they appear to deliver better results in tasks such as code generation, processing of long documents, and building agents that take actions independently.
read also
OpenAI wants to remove largest AI model GPT-4.5 from API
Companies such as Windsurf, Qodo, Blue J, Hex, and Thomson Reuters tested GPT-4.1 in production. They report improvements in efficiency, accuracy, and usability for complex tasks, such as legal analysis or code review.
GPT-4.1 is exclusively available through the API. In ChatGPT, improvements from GPT-4.1 continue to be gradually integrated into the GPT-4o version.