OPENAI launched a New family of AI models This morning, which considerably improves coding capacities while reducing costs, directly responding to growing competition on the company’s AI market.
The IA company based in San Francisco introduced three models-GPT-4.1, GPT-4.1 Mini and GPT-4.1 Nano-All Available immediately through his API. The new range works better in software engineering tasks, follows instructions more precisely and can treat up to a million context tokens, which is equivalent to around 750,000 words.
“GPT-4.1 Offers exceptional performance at a lower cost,” said Kevin Weil, Product Director in Openai, during Monday’s announcement. “These models are better than GPT-4O on almost all dimensions.”
The price may be the most important for corporate customers: GPT-4.1 will cost 26% less than its predecessor, while the Nano light version becomes the most affordable offer of Openai at only 12 cents per million tokens.
How GPT-4.1 improvements target the biggest points of pain in business developers
In a Candide interview with Venturebeat, Michelle Pokrass, research manager after training in Openai, stressed that practical commercial applications have led the development process.
“GPT-4.1 was formed with a single objective: to be useful for developers,” Pokrass told Venturebeat. “We have found that GPT-4.1 is much better to follow the types of instructions that companies use in practice, which facilitates the deployment of ready for production.”
This focus on the real world utility is reflected in the reference results. On Swe-Bench checkedwhich measures software engineering capacities, GPT-4.1 marked 54.6%-a substantial improvement of 21.4 points of point compared to GPT-4O.
For companies that develop AI agents who work independently on complex tasks, improvements in the following teaching are particularly precious. On the Multi-Calculist reference of Scale, GPT-4.1 marked 38.3%, outperforming GPT-4O of 10.5 percentage points.
Why the strategy of the three -level openai model questions competitors like Google and Anthropic
The introduction of three distinct models at different prices addresses the diversifying AI market. The GPT-4.1 Lightlights complex corporate applications, while the Mini and Nano versions approach use cases where speed and profitability are priorities.
“All tasks do not need the most intelligent or most important capacities,” Pokrass told Venturebeat. “Nano will be a model of battle horse for use cases such as semi-automatic entry, classification, extraction of data or anything else where speed is major concern.”
Simultaneously, Openai has announced its intention to depreciate Overview GPT-4.5 – His largest and most expensive model published only two months ago – of his API before July 14. The company has positioned GPT-4.1 As a more profitable replacement which offers “improved or similar performance on many key capacities at a much lower cost and latency”.
This decision allows OPENAI to recover IT resources while providing developers with a more effective alternative to its most expensive offer, which had been at the price of $ 75 per million entry tokens and $ 150 per million exit tokens.
Real world results: how Thomson Reuters, Carlyle and Windsurf take advantage of GPT-4.1
Several corporate customers who have tested the models before launch have reported substantial improvements in their specific areas.
Thomson Reuters experienced an improvement of 17% of the precision of the multi-documentary examination when using GPT-4.1 with its AI legal assistant, Coconsel. This improvement is particularly useful for complex legal work flows involving long documents with nuanced relationships between clauses.
Financial company Carlyle Reported 50% of better performance on the extraction of granular financial data from dense documents – a critical capacity for investment analysis and decision -making.
Varun Mohan, CEO of the coding tool provider Windsurfing (formerly Codeium), shared detailed performance measures during the announcement.
“We have found that GPT-4.1 reduces the number of times it needs to read unnecessary files of 40% compared to other leading models, and also changes unnecessary files 70% less,” said Mohan. “The model is also surprisingly less verbose … GPT-4.1 is 50% less verbose than other leading models.”
Context to one million things: what companies can do with 8 times more treatment capacity
The three models include a context window of a million tokens – eight times larger than the token limit of 128,000 token from GPT -4O. This enlarged capacity allows models to process both several long documents or entire code bases.
In a demonstration, Openai showed GPT-4.1 Analyzing a NASA server journal file of 450,000 people from 1995, identifying an abnormal input deeply hiding in the data. This capacity is particularly precious for tasks involving large sets of data, such as code standards or corporate documents.
However, Openai recognizes the degradation of performance with extremely important entries. On his internal OPENAI-MRCR testPrecision rose from around 84% with 8,000 tokens to 50% with a million tokens.
How the company’s AI landscape moves while Google, Anthropic and Openai compete for developers
The exit comes as the competition in the Enterprise Ai Space warms up. Google recently launched Gemini 2.5 Pro with a comparable context window of a million, while anthropic Claude 3.7 SONNET gained ground with companies looking for alternatives to Openai offers.
The Chinese AI startup Deepseek has also recently improved its models, putting additional pressure on Openai to maintain its leadership position.
“It was really cool to see how improvements in understanding the long -term context resulted in better performance on specific verticals such as legal analysis and extraction of financial data,” said Pokrass. “We have found that it is essential to test our models beyond academic references and ensure that they work well with companies and developers.”
By releasing these models specifically through its API Rather than Chatgpt, Openai reports its commitment to developers and corporate customers. The company plans to gradually integrate the features of GPT-4.1 to Chatgpt over time, but the main objective remains on the supply of robust tools for companies that create specialized applications.
To encourage additional research on long context processing, Openai publishes two assessment data sets: OPENAI-MRCR to test the coreference capacities of several rounds and Chart To assess the complex reasoning on long documents.
For corporate decision -makers, the GPT-4.1 family Offers a more practical and profitable approach to the implementation of AI. While organizations continue to integrate AI into their operations, these improvements in reliability, specificity and efficiency could accelerate adoption in industries always weighing the costs of implementation compared to potential advantages.
While competitors are pursuing larger and more expensive models, OpenAi’s strategic pivot with GPT-4.1 suggests that the future of AI may not belong to the most important, but most effective models. The real breakthrough is perhaps not in references, but by providing an AI of business quality within the reach of more companies than ever.
