OpenAI announced the launch of three new artificial intelligence (AI) models on Monday, branded as the GPT-4.1 family. This latest iteration includes three versions: GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. The San Francisco-based company stated that these models surpass the performance of their predecessors, GPT-4o and GPT-4o mini, in several key areas, particularly in coding capabilities and instruction adherence. Importantly, these advanced large language models (LLMs) are exclusively accessible to developers through an application programming interface (API), and will not be offered via ChatGPT.
OpenAI Unveils Coding-Centric GPT-4.1 AI Models
In a recent blog post, OpenAI announced the introduction of the GPT-4.1 series, which boasts several enhancements compared to the GPT-4o models. Key improvements include heightened performance, an expanded context window, and more competitive pricing. The knowledge base for these models has also been updated to include information available until June 2024.
Designed specifically for tasks related to coding and activities requiring better adherence to instructions, OpenAI highlighted that the GPT-4.1 model achieved a score of 54.6 percent on the SWE-bench Verified benchmark. This score not only surpasses that of GPT-4o but also exceeds the recently launched GPT-4.5, positioning GPT-4.1 as the top choice for coding tasks.
Additionally, the GPT-4.1 model attained a score of 38.3 percent on the MultiChallenge benchmark related to instruction following and registered a 72 percent score on the Video-MME benchmark, which evaluates multimodal long-context understanding. All models in this series feature a context window capable of handling up to one million tokens.
Overview of OpenAI’s GPT 4.1 AI models
Photo Credit: OpenAI
While GPT-4.1 serves as the flagship model of the series, the GPT-4.1 mini also demonstrates superior performance against GPT-4o across various benchmarks, including intelligence assessments. OpenAI noted that the mini variant achieves latency reduction by half and offers an 83 percent lower cost compared to the GPT-4o model.
Furthermore, the company is promoting the GPT-4.1 nano as ideal for tasks requiring low latency. It boasts an 80.1 percent score on the MMLU benchmark, a 50.3 percent score on GPQA, and a 9.8 percent score on the Aider polyglot coding benchmark, indicating its performance surpasses that of GPT-4o mini. OpenAI emphasized that these models’ enhanced instruction-following capabilities and extensive context window make them suitable for powering AI agents.
Regarding pricing, GPT-4.1 is set at $2 (approximately Rs. 171) per million input tokens and $8 (approximately Rs. 685) per million output tokens. The mini variant is priced at $0.40 (around Rs. 34) per million input tokens and $1.60 (approximately Rs. 137) per million output tokens, while the nano model offers the most economical option at $0.10 (roughly Rs. 8.5) per million input tokens and $0.40 (about Rs. 34) per million output tokens.
The GPT-4.1 series will only be available to developers through API access. Nonetheless, OpenAI has indicated that the enhancements introduced in these models will be integrated into the newest version of GPT-4o, which is currently available in ChatGPT.
