Maison AI Integrates Next-Generation AI Model "GPT-5.4" Series, Completely Revamping Conventional GPT-5 Systems for Enhanced "Speed" and "High-Precision" Utilization

AuthenticAI's Maison AI platform has undergone a major update, integrating the next-generation "GPT-5.4" series of AI models. This revamp of the conventional GPT-5 systems offers four optimized lineups, from "nano" to "inference-specialized," enabling faster and more accurate AI utilization tailored to specific business objectives and difficulty levels, minimizing waste.
product_launchNQ 58/100出典:PR Times

📋 Article Processing Timeline

  • 📰 Published: April 3, 2026 at 00:00
  • 🔍 Collected: April 2, 2026 at 19:36
  • 🤖 AI Analyzed: April 17, 2026 at 23:44 (364h 7m after Collected)

AuthenticAI Inc. (Headquarters: Minato-ku, Tokyo; Representative Director: Toru Ueda; hereinafter "AuthenticAI"), a developer and provider of enterprise generative AI platforms, has implemented a major update to its generative AI platform "Maison AI." We are pleased to announce that this update has completely revamped the conventional GPT-5 series models, introducing the higher-performance and optimized next-generation "GPT-5.4" series (four lineups).

AuthenticAI aims to realize a society where everyone can naturally master generative AI, and provides the enterprise generative AI platform "Maison AI." Beyond merely providing tools, we promote the transformation of our customers into "AI-native companies" through "accompanying support" that draws out corporate creativity and strategy through generative AI.

■ Background: The era of AI is shifting from "using" to "choosing" according to purpose

As the social implementation of generative AI rapidly progresses, on-site users are increasingly expressing concrete needs tailored to practical work, such as "I want to get answers faster," "I want it to think more deeply about complex problems," and "I want to reduce the waste of using an excessive model for the difficulty of the task."

This revamp to the "GPT-5.4" series responds to these expectations for both "speed" and "accuracy," and "efficiency through optimal model selection." With four optimized lineups based on conventional models, it delivers the highest level of performance in all business scenarios, from daily chats to advanced data analysis. (※1)

Comparison of GPT‑5.4 mini / nano and GPT‑5 mini

(※1)Quoted from OpenAI's announcement "Introducing GPT-5.4 mini and nano"

https://openai.com/ja-JP/index/introducing-gpt-5-4-mini-and-nano/

■ Proven, overwhelming "practical execution capability" in major benchmarks

The latest "GPT-5.4" goes beyond mere text generation, demonstrating its performance as an "agent-based AI" capable of autonomous decision-making and complex operations in various benchmarks.

Computer operation performance surpassing humans:

In "OSWorld-Verified," which measures the ability to directly operate a desktop environment and autonomously execute workflows, GPT-5.4 recorded a success rate of 75.0%. This result surpasses the average human performance (72.4%), demonstrating that AI can act as a proxy for advanced business processes. (※2)

Deepening of engineering and logical reasoning:

It achieved 75.1% in "Terminal-Bench 2.0," which measures software engineering capabilities, and 91% accuracy in "BigLaw Bench" for the legal domain. Factual errors (hallucinations) have been reduced by approximately 33% compared to conventional models, enabling highly reliable practical support. (※3)

Vast contextual understanding of 1 million tokens:

Supports a context window of up to 1,050,000 tokens. It can analyze several years of financial reports or hundreds of pages of legal contracts in bulk, performing highly accurate inference while maintaining information consistency.

*In Maison AI, the maximum number of characters that can be remembered in a single conversation is 360,000 characters.


(※2)Quoted from OpenAI's announcement "Introducing GPT-5.4"
https://openai.com/ja-JP/index/introducing-gpt-5-4/

(※3)Quoted from OpenAI's announcement "Introducing GPT-5.4 mini and nano"

https://openai.com/ja-JP/index/introducing-gpt-5-4-mini-and-nano/

■ New Generation "GPT-5.4" Series: Four Lineups

① High-speed: GPT-5.4-nano
An ultra-lightweight model boasting astonishing response speed. Ideal for speed-priority tasks such as large-scale data sorting and information extraction.

② Standard: GPT-5.4-mini
A medium-sized model with an extremely excellent balance of speed and accuracy. It offers performance close to that of higher-end models in software engineering metrics (SWE-bench Pro) in a cost-effective manner.

③ Standard: GPT-5.4
A frontier model with the most balanced capabilities for writing, coding, and tool use. Significantly reduced factual errors make it function as the core of professional practical work.

④ Inference: GPT-5.4 (for inference)
The pinnacle model allocated with computational resources for "thinking deeper and longer." It addresses challenges requiring advanced logical reasoning, such as scientific hypothesis testing and complex data analysis.

■ New workflows brought by "faster, higher performance"

This revamp allows users to select the optimal "intelligence" for their specific tasks. By enabling model selection that matches purpose and means—such as "I need speed now, so I'll use nano" or "This analysis is complex, so I'll leave it to the inference model"—it reduces the stress of AI utilization and maximizes time for creative work.

Maison AI will continue to rapidly incorporate the latest generative AI technologies into its platform, evolving further as an "AI transformation partner" supporting corporate change.

For more details on this update, please visit the following page:
https://maisonai.io/blogs/update/ver1-87

■ Main functions and usage scenarios

AuthenticAI will continue to rapidly incorporate the latest AI models and support AX (AI Transformation) for companies in all industries. We will accelerate the creation of new value born from human-AI co-creation and pioneer the frontier of AI agent technology.


■ Generative AI Platform "Maison AI" Service Overview

"Maison AI" provided by AuthenticAI Inc. is a generative AI platform specialized for creative industries such as fashion, interior design, and miscellaneous goods. It is equipped with 14 types of high-performance generative AI models, including GPT-5 and Gemini 3 Pro, demonstrating high performance in both language understanding and image generation. Furthermore, it features a multifunctional toolset including AI agents tailored for various specialized professions such as designers and merchandisers, enabling utilization for diverse needs from operational efficiency improvement to new business development and educational applications.

"Maison AI" Official Service Website: https://maisonai.io/

■ About AuthenticAI Inc.

AuthenticAI Inc.
Address: World Kita-Aoyama Bldg. 12F, 3-5-10 Kita-Aoyama, Minato-ku, Tokyo
Business: Provision of solutions and products utilizing generative AI
Established: November 8, 2024
Representative Director: Toru Ueda
Official Website: https://authenticai.co.jp/

■ Inquiries

If you have any questions regarding this matter, please contact us using the inquiry form below.
Inquiry Form: https://authenticai.co.jp/pages/contact

FAQ

What is the GPT-5.4 series?

It is the next-generation AI model integrated into Maison AI by AuthenticAI, completely revamping the previous GPT-5 series with significantly improved speed and accuracy. It offers four lineups to choose from based on business objectives and difficulty.

What types of tasks are Maison AI's new models best suited for?

'nano' is ideal for high-speed data sorting and information extraction, 'mini' for tasks requiring a balance of speed and accuracy, 'GPT-5.4' for writing and coding, and the 'inference-focused' model for scientific hypothesis testing and complex data analysis.

What are the main features of the GPT-5.4 series?

Key features include computer operation performance surpassing human capabilities, deepened engineering and logical reasoning, and extensive contextual understanding of over 1 million tokens. Factual errors have also been reduced by approximately 33%.