OpenAI Releases GPT-4.1 Family of AI Models in API, Said to Outperform GPT-4o

OpenAI launched three new synthetic intelligence (AI) fashions on Monday. Staying true to its uncommon naming conference, the corporate known as this the GPT-4.1 household of AI fashions. There are three variants — GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. The San Francisco-based AI agency mentioned that these fashions outperform the GPT-4o and GPT-4o mini throughout a number of metrics, with a big enchancment in coding and instruction following. Notably, these new giant language fashions (LLMs) are solely obtainable to builders through an utility programming interface (API), and won’t be accessible through ChatGPT.

OpenAI Releases Coding-Focused GPT-4.1 AI Models

In a publish, OpenAI introduced the discharge of the GPT-4.1 sequence of AI fashions. These API-exclusive fashions get a number of upgrades over the GPT-4o fashions, together with larger efficiency, elevated context window, and improved pricing. The data cutoff of these fashions has additionally been upgraded to June 2024.

OpenAI mentioned that these fashions have been designed for coding-related duties and actions that require larger instruction adherence. Highlighting the coding proficiency, OpenAI claimed GPT-4.1 scored 54.6 p.c on the SWE-bench Verified. This rating is larger than each GPT-4o and the just lately launched GPT-4.5, making 4.1 the corporate’s main mannequin for coding.

GPT-4.1 additionally scored 38.3 p.c on the MultiChallenge benchmark for instruction following, and 72 p.c on the Video-MME benchmark for multimodal lengthy context understanding, the corporate claimed. Notably, all of the AI fashions in the sequence have a context window of up to a million tokens.

Overview of OpenAI’s GPT 4.1 AI fashions
Photo Credit: OpenAI

While GPT-4.1 stays the premium mannequin of this sequence, the GPT-4.1 mini can also be claimed to outperform GPT-4o in a number of benchmarks, together with intelligence evaluations. OpenAI mentioned the mini mannequin reduces latency by half and prices 83 p.c much less in contrast to 4o.

The AI agency can also be pitching the GPT-4.1 nano as one of the best mannequin for low latency duties. With an 80.1 p.c rating on MMLU, 50.3 p.c rating on GPQA, and 9.8 p.c on the Aider polyglot coding, OpenAI says the AI mannequin fares higher than the GPT-4o mini mannequin. OpenAI additionally highlighted that the improved instruction following and the lengthy context window additionally make these fashions an choice in powering AI brokers.

Coming to pricing, the GPT-4.1 prices $2 (roughly Rs. 171) per million enter tokens and $8 (roughly Rs. 685) per million output tokens. The mini variant is priced at $0.40 (roughly Rs. 34) per million enter tokens and $1.60 (roughly Rs. 137) per million output tokens. And, the most affordable nano mannequin will price builders $0.10 (roughly Rs. 8.5) per million enter tokens and $0.40 (roughly Rs. 34) per million output tokens.

Notably, the GPT-4.1 sequence of AI fashions will solely be obtainable through API to builders. However, the AI agency acknowledged that the enhancements of these fashions can be transitioned into the most recent model of GPT-4o, which is obtainable in ChatGPT.