OpenAI’s new GPT-4.1 AI models focus on coding

On Monday, Openai launched a new family of new models called GPT-4.1. Yes, “4.1” – as if the company’s nomenclature was no longer confused.

GPT-4.1, GPT-4.1 mini and GPT-4.1 Nano, all of this says “Excel” in Openai coding and guidelines. Openai’s API is not available ChatgptMultimodal models have 1 million-million-million taxi-context windows, ie in about 750,000 words (longer than war and peace).

GPT-4.1 increases efforts to create advanced programming models such as Openai competitors like Google and anthropic. Google’s recently released Gemini 2.5 ProWith 1 million Token Context window, popular coding criteria are highly high. So anthropic Claude 3.7 Sonnet and Chinese AI Startup Deepseek’s Improved V3.

This is to cultivate AI coding models, including many technological giants, including complex software engineering responsibilities. Openai’s Grand Ambition is to create an “agent software engineer” CFO Sarah Friar put him During a technological summit in London last month. The company claims that future models will be able to end all applications to end, quality assurance, error testing and writing documents.

GPT-4.1 is a step in this direction.

“Formats, consistent tools and more formats, consistent tools and more formats, consistent tool usage and more formats, consistent tool usage and more formats, which are the use of a consistent tool, and more, allow these adjustments to build many improvements in real-world engineering responsibilities.”

Openai clearly claims the full GPT-4.1 model GPT-4O and GPT-4O Mini Models on coding criteria including swe-bench. GPT-4.1 Mini and Nano’s price is more efficient and faster than the price of accuracy and faster, GPT-4.1 Nano with Openai is the fastest – and the cheapest – model.

GPT-4.1 costs a million signs and millions of millions of dollars. GPT-4.1 Mini, 0.40 dollars / million dollars and $ 1.60 / million in exit tokens and GPT-4.1 Nano, $ 0.10 / million dollars and $ 0.40 / million dollar access tokens.

According to Openai’s domestic test, GPT-4O (32,768) can provide more verses (32,768) more than 52% compared to 16,384, between 52% and 54.6% in the bottom of the Swe-bench. (Openai, SWE-DEN Solutions The infrastructure of the problems, so it was noted that the scores could not work in the range.

In a separate assessment, Openai, using video-MME using a video-MME designed to measure the ability to “understand” the content of a model in the video. GPT-4.1 has reached 72% accuracy in 72% of the “Long, Subtitles” video category.

GPT-4.1 is important in Benchmarks and the latest “Knowledge Cut” in the latest “knowledge” (by 2024), the specialists of some of the best models today are important. For example, too work make rich honest These code-generating models are often not corrected and cannot even provide security vulnerabilities and errors.

Openai, GPT-4.1 is more reliable (ie to make a mistake) more reliable (ie to make mistakes), Openai-Mrack reduced to 50% from 84% to 80% to 1 million token. GPT-4.1 also tends to be more “literal”, at the same time, sometimes requires more specific, open instructions.