The new pattern of the VPTale’s VPTs of AI

Open Monday launched a new family of models called Gpt-4.1. Yes, “4.1” – as the company’s nomenclature was not confused already.

There is the Gpt-4.1, Gpt-4.1 mini of the Gpt-4.1 Nano, all of the “Excel” coding and instruction. Available through the apien’s apes but not ChartMultimodal models have multimo-token-token-token context, which may take place in about 750,000 words in a vault (longer than “war”).

The Gpt-4.1 Arrives as rivals Open as Google and the rearing anthrophy to build sophisticated programming patterns. Geop is recently released Gemini 2.5 prothat also has a context of the context of 1 millionaire, highly ranks on popular codifying benchmarks. So make the anthropic Claudius 3.7 sonnet and the Chinese UI’s start Updated update of Deepseek. I am

It is the object of many technological giants, including Opena, to form modeling modeling achievement of complex software engineering. The great ambition of Openai is to create a “agent” software engineer, as Cfo sarah friar During a top of the technology in London last month. The company is assisting the future patterns will be allowed all approves at the end of ending, of Illery Hands, the quality test, bug test, and documentation.

The Gpt-4.1 is one step in this direction.

“We optimized Vpt-4.1 for the world’s use of the best of the shores, and more” a tool, and more “a tool, and more” a device that are considerably better to the jobs in the software in engineering in the software of the software. “

Open I pretend the Gpt-4.1 Gpt-model Gpt-4o and gpt-4O Models on coding benchmarks including Sw-Bench Bank. Gpt-4.1 mini and nano they say to be more efficient and faster, with an accuracy, with operations by saying Gpt-4.1 Nano is its faster – model.

The Gpt-4.1 costs $ 2 per millions of entry tokens and $ 8 per millions of output tokens. Gpt-4.1 mini 0.40 / m input tokens and $ 9.60 / m Out-4.1 nano is $ 0.10 / m input tokens.

According to the internal proof of Opena, Gpt-4.1, which may generate more tokens in the VPT-4O (32.768 versus, held between 52% and 54 € on the Sw-bace. (Open Nutated in a Blog since some SW-Bencha Troubles could not stop on his infrastructure, stationary of the parties.) Those equals Gemini 2.5 priest 3.7

In a separate evaluation, Openiya probably has a square using video-mme, which is designed to measure the ability to “understand” content in videos. The VPTU-4.1 reached a card cartoon 72% on the “long, no Settites” Categories Video, Openi claims.

While getting raised in reasonably well in benchmarks and has a more recent cock “, harm a better task of events that you are not expert entrants. For example, so many Studies to get shown May the generative models of the code often fail, and also present, vulnerability and bugs.

Open recognizes, that, that Gpt-4.1 becomes less reliable (ie as mistakes) the most input tokens should handle. In one of the proof of the company, open-micr, the model accuracy declined by 84% with 8,000 tokens to 50% with 1 million tokens. The Gpt-4.1 also tender to be more “literal” of the Vpt-4o, says the company, sometimes requires more specific, explicit.