On Monday, Openai started a new model family called GPT-4.1. Yes, “4.1” – Company Nomenclature is not enough.
GPT-4.1, UPT-4.1 Mini and GPT-4.1 Nano say “Excel”. Outai's API can be available through ChatgptIn the multimodal models, 1 million words that have 1 million-year-olds in 1 million token, with about 1 million words of 750,000 words (“war and peace”.
GPT-4.1 arrives Arpirav competitors, such as Google and Mae Ropet Ratchet Up Efforts to set up Google and Anthropic RatChattcodes. Google's recent released Gemini 2.5 ProIt has a window in more than 1 million token conditions and has a high level of popular coding benchmarks. So do anthropic's Claude 3.7 Sonnet And Chinese AI startup Deepseek's upgrade v3.
It has been able to perform the goal of technology for technology companies, including Openai, including Openai, including Openai, operational, including Openai, including Openai. Openai's goal is to create “Agentic Software Engineer” Cfo Sarah Friri put it During the technology summit last month in London. The company has to end its future models the entire app and quality assurance,
GPT-4.1 is a step in this direction.
“To order the developer construnching of the Frontors and Order,” Developers enable developers to build a significant agent in real software engineers. “
Openai shows its full of its full GPT-4O and GPT-4O Mini Mini Models on Coding Standards including Swe-Bench. The GPT-4.1 Mini and Nano are said to be fasterly faster in some of the costs of some accuracy. GPT-4.1 Nano is its fastest and cheapest.
GPT-4.1 costs $ 8 per million in $ 2 million and output tokens per million. The GPT-4.1 mini is $ 0.40 / million and $ 1.60 / million output tokens and GPT-4.1 Nano and GPT-4.1 Nano and $ 0.40 / million.
GPT-4.1 (32,768) compared to 32,768 comparisons (32,768 comparisons) (32,766) 16,384 (34,766) compared to cart (32,766) 16,384 (32,766) compared to 32,766 (32,766) compared to 16,384. (These figures in Openai in Openai in Openai, which are in Openai, who are in the blogging post in the blogging post, which can solve the problems that will solve the problems of the problems of the problem with the problem of solutions, are provided under the provisions.
Openai using the GPL-4.1 video MME in a separate assessment. In the video video of the model, the video -me designed to understand “understand”. GPT-4.1 obtained Table-72% accuracy in the “long, steps” video category.
GPT-4.1 and better “knowledge cuts” for basic events has been a better reference to the current “vulnerable cuts” for current events. For example, Many Studies Have Reveal These code-generate models often fail to repair;
Openai, GPT-4.1 is less reliable (meaning to make mistakes) increases input tokens. In one of the company's self-test, Openai-MRCR decreased by 84% to 8,000 to sky with 1 million tokyos from 84 to 8,000 tokyos. GPT-4.1 is often “literal” than “literal” than GPT-4o.