Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Open one initialized a Ai’s new family of AI models This morning, when cutting costs, the ability to significantly improve coding abilities, responds to the competitive competition in the AI market.
San Francisco-based AI company presented three models – GPT-4.1, GPT-4.1 Mini and GPT-4.1 Nano – All Available immediately Through the API. The new drawing program fulfills software engineering instructions, and can process the instructions and process up to a million feet, equal to 750,000 words.
“GPT-4.1 offers exceptional performance at a lower price,” said Kevin Weil, Openai, during Monday’s announcement. “These models are better than only GPT-4O in each size.”
Perhaps the most important prices for enterprise customers: GPT-4.1 It is 26% less than its predecessor, a light nano version, Openai’s most favorable offer, which offers the most favorable in only 12 kopecks for one million tokens.
Improving GPT-4.1 The biggest pain points of target enterprise developers
Michel Pokrass with VentureBeat, Michel Pokrass, Michel Pokrass, stressed that practical work applications are under development.
“GPT-4.1 has been trained for a purpose: be useful for developers,” Pokrass said to Venturebeat. “We found GPT-4.1 that we found the types of instructions used in the practice, which makes it easier to place production.”
This focus on the real-world program is reflected in the results of benchmark. Side Swe-bucketSoftware measuring engineering opportunities, GPT-4.1, 54.6%, 21.4% point over GPT-4O.
To develop an AI agents that work independently in complex tasks, the progress in the following guidelines is especially valuable. GPT-4.1 in the size of the size of MultiMhallenge 38.3%, GPT-4o preferred the 10.5 percent statement.
Why Openai’s three-speed model strategy calls opponents like Google and anthropic
Application of three different models in different price points appeals to the Diversification of the AI market. Flaghip GPT-4.1 uses cases used in cases where speed and cost efficiency in mini and nano versions are priority.
“All tasks do not need the most intelligence or top opportunities,” Pokrass said to Venturebeat. “Nano, autocomplete, classifications, classifications, information removal or velocity will be a work model for use of works like anything else.”
At the same time, Openai announced plans to wear GPT-4.5 preview – The largest and most expensive model released only two months ago – from API to 14 July. GPT-4.1 As a more efficient replacement that provides “improved or similar performance” in many key opportunities in a lower price and delay.
This movement allows developers to reconsider the cost of $ 75 in Tokens worth $ 75 and millions of dollars in $ 75, while providing the most effective proposal for $ 75.
Real-world results: Thomson Reuters, Carlyle and Windsurf use GPT-4.1
Several enterprises, who tested the models before the beginning, made significant improvements in their unique domains.
Thomson Reuters When using the Legal AI Assistant GPT-4.1, I saw a 17% improvement in multiple documentary review accuracy, Coqueted. This expansion is especially valuable for the complex legal work that covers long documents with the Nuhance relations between the dams.
Financial firm Carlyle 50% better performance to withdraw granular financial information from intense documents – a critical ability for investment analysis and decision making.
Varun Mohan, CEO with a code of coding Wind (Previously codium), shared detailed performance sizes during the announcement.
“GPT-4.1 reduces the number of unnecessary files to read 40% compared to other leading models and changes unnecessary files to 70% less,” he said. “The model is also a surprisingly low venture … GPT-4.1 is 50% less verbais than other leading models.”
Million-token context: 8x which businesses can do with more processing capability
All three models are a context window of one million Token – 128,000 Token of GPT-4o to 128,000 Token. This allows you to process multiple documents or all codes in the expanded capacity models.
In a demonstration, Openai showed GPT-4.1 that determines the deeply hiding the input document in 450,000 token NASA server login document since 1995. It is especially valuable for tasks related to large data staff, such as abilities, code deposits or corporate document collections.
However, Openai accepts performance with extremely large entrances. In the interior Openai-Mrcr testAccuracy was reduced by 84% to 84%, up to 50% of one million Token.
How does the landscape of the enterprise change as Google, anthropic and Openai developers
The free release comes as competition in the enterprise heating the AI space. Google has recently been launched Gemini 2.5 Pro A million-token context window that can be comparable, although anthropic Claude 3.7 Sonnet Openai’s proposals were tacked with an alternative institutions.
China’s EU Beginner DeepSeek has recently improved the models by showing additional pressure to Openai to protect the leadership position.
“How to progress better performance in concrete skools, as a long context exacerbate, legal analysis and financial information, it was very nice,” he said. “We saw our models are very important to try out of academic criteria and make sure that it works well with enterprises and developers.”
By leaving these models on their own Vibrate More than Chatgpt, Openai, developers and enterprise customers have signed their loyalty. The company plans to unite the features of GPT-4.1 in time, but the initial attention remains to ensure solid tools for the establishment of specialized applications.
To promote another study in long cookie production, Openai releases two assessment databases: Openai-MRCR To try the abilities of many round coreefere and Graphs In order to assess complex thinking among long documents.
For enterprise decisions GPT-4.1 Family Offers a more practical, efficient approach to AI application. As organizations continue to integrate the AI, the cost of reliability, specificity and efficiency, the cost of implementing potential benefits can still accelerate the adoption within the expansion costs.
While competitors follow the bigger, Boster models, while following the Strategic Pivot of GPT-4.1, the EU’s future may not belong to the greatest models, but it may belong to the most effective ones. True improvements may not be in benchmarks, but in bringing the EI AI in the reach of the enterprise more than before.
Source link
