While we wait for Openai to finally display his new “Open Source” model, he has just announced the new GPT-4.1. Its great objective is the program and was born to counteract the thrust of Gemini 2.5 Pro from Google, which in a few weeks has expanded as the foam, and not only for its capabilities for programming, but for creative reading.
It is a way to return the blow that Gemini 2.5 Pro that offers a context window of 1 million tokens and has been able to demonstrate great performance in the most popular programming benchmarks. And we say that it was born that way because, if Openai’s proposal for a couple of months ago was to reduce all its models to an upcoming GPT-5, now life is complicated again with a new nomenclature.
There is a GPT-4.1, GPT-4.1 Mini and GPT-4.1 Nano And all, according to Openai, are characterized by their ability to program and monitor orders. Multimodal models have a context window of one million tokens in order to put at the same height as Gemini 2.5 Pro that offers the same capacity.
Just yesterday OpenAi declared that GPT-4.1 becomes the perfect substitute for GPT-4.5, While opening access to their API to developers after it was released as “Orion” in February as a response to the tsunami that was Deepseek at the end of January. GPT-4.1 offers a performance similar to GPT-4.5.
The only thing there is a fairly important difference against Google’s AI, for now the three GPT-4.1 models are available only through the API of Openai and not from Chatgpt. Gemini 2.5 Pro is available in Spain through the Advanced subscription, so at this point it plays with great advantage, although the approach of this type of models goes more for developers.
Precision verified by Swe-Bench
The free android
And not only has to face Gemini 2.5 Pro, but also Claude 3.7 Sonnet and Deepseek, so the arrival of GPT-4.1 Open a large range of options For any user or developer. In fact, the objective of all technological giants is to be able to launch models of programming that are capable of performing complex software engineering tasks.
It is precisely where Openai is going, being able to create An agricultural software engineer (Word that has already used adobe days ago to refer to AI agents) and that future models are able to program entire apps to end, manage aspects such as quality tests, bugs search and create the necessary documentation.
The new OpenAI model exceeds GPT-4O and GPT-4o Mini In the programming tests published from Your websiteand GPT-4.1 Mini and Nano are more efficient and faster at cost in some precision in their results. Nano is the fastest and cheap model ever released by OpenAI as collected from his own words.
OpenAi has given all the details of GPT-4.1 to Techcrunch: “We have optimized GPT-4.1 for real use based on feedback to improve areas in which developers are more careful: programming Front-endmake less extra editions, consistent use of tools, the monitoring of sustainable formats and more. “
In the tests It was below the scores reported by Google and Anthropic For Gemini 2.5 Pro (63.8 %) and Claude 3.7 Sonnet (62.3 %) with GPT-4.1 marking between 52 and 54.6 %in Swe-Bench Verified. Anyway, even if these models are dedicated to programming, it still remains to be perfect, since sometimes They generate failures that transform into vulnerabilities.
