#Japan #Chiyoda #FlashLabs #OrcaRouter #Qwen3.8

Introduction

The AI inference gateway named OrcaRouter, exclusively offered by FlashLabs in Japan, has made a significant impact by securing top positions in the Hugging Face rankings. Its underlying open model, Qwen3.8-27B-Uncensored, developed by the American AI research organization Continuum AI, is now ranked 3rd globally among over three million models as of August 21, 2026, according to company data. Notably, the same series also saw its FP8 version ranked 6th, marking a significant achievement for FlashLabs.

Key Highlights

• - The model Qwen3.8-27B-Uncensored-MLX achieved 3rd place on Hugging Face.
• - The version Qwen3.8-27B-Uncensored-FP8 came in 6th place, representing two models from the same series in the top 10 simultaneously.
• - The FP8 version garnered over 100,000 downloads within just six days of its release.
• - The primary focus of these models is cyber security research and red teaming, leading to restricted access for users.

Model Rankings

The current ranking reflects a rapid ascent for FP8, which entered the top ten just four days after its release on August 15, 2026. The subsequent update on August 21 confirmed that the MLX version ranked 3rd, while the FP8 version claimed 6th place. This remarkable feat indicates growing interest and usage within the AI security research community, evidenced by the swift downloads for the FP8 model.

Development Background

Historically, cyber security research has required models that can circumvent standard safety mechanisms. Continuum AI, recognizing the limitations within conventional models, sought to address these challenges by developing the Uncensored version based on the Qwen3.8-27B architecture. This model effectively reduces refuse responses using a method termed Abbreviation— a pioneering approach that seeks to enhance responsiveness without compromising safety controls.

Multi-Modal Capabilities

The Qwen3.8-27B-Uncensored model supports up to 262K tokens of long context and allows for diverse inputs, including text, images, and videos. Its six available formats include:

• - BF16: High-precision version maintaining baseline model accuracy.
• - FP8: An offline block quantized version, ensuring compatibility with existing infrastructures.
• - GGUF: For local execution, supporting various bit quantization.
• - MLX: Designed for Macintosh environments with specific requirements.
• - INT8: A standardized quantization format widely supported by major inference engines.
• - NVFP4: A format optimized for high-speed inference on NVIDIA's latest GPUs.

Restrictive Access

While the aforementioned features are available, Access to the OrcaRouter API is currently restricted to large enterprises engaged in cyber security operations. The general public’s access is gradually being phased out to ensure that the models are employed responsibly. However, the standard Qwen3.8-27B model continues to be offered at no cost, allowing developers to start validation and testing immediately.

Looking Ahead

Moving forward, FlashLabs intends to maintain its momentum in supporting AI security research, contributing to the establishment of safer AI utilization environments by continually offering innovative models and infrastructures.

About OrcaRouter

OrcaRouter serves as an advanced AI inference gateway developed by Continuum AI, with exclusive delivery in the Japanese market by FlashLabs. It features compatibility with over 200 language models and generative AI frameworks through an OpenAI-compatible API, providing unprecedented access to cutting-edge technology.

Company Overview

FlashLabs, based in Chiyoda, Tokyo, focuses on AI application research aimed at automating customer experience through a human-AI hybrid approach. With a commitment to innovation and excellence, FlashLabs is positioned to reshape the landscape of AI applications in Japan and beyond. For further inquiries, please contact:
FlashLabs Marketing Department
Email: [email protected]

This article highlights the remarkable achievements of the OrcaRouter and its continuing journey in advancing AI technology for secure and responsible use.