FlashLabs releases uncensored DeepSeek-V4 multimodal model for security research

FlashLabs has announced the release of the GGUF quantized version of ‘DeepSeek-V4-Flash-Vision-Uncensored’ via its OrcaRouter AI inference gateway. Developed by Continuum AI and distributed exclusively by FlashLabs in Japan, this experimental multimodal model is designed specifically for security researchers, including red and blue teams.

Based on the ‘DeepSeek-V4-Flash-Vision-Exp’ model, this version utilizes abliteration to remove refusal responses directly from the model weights, rather than relying on jailbreak prompts. The MXFP4 quantized version, which is approximately 156GB in size, supports a context window of up to one million tokens and is available on Hugging Face under an MIT license.

In related developments, the OrcaRouter Model Leaderboard has updated its rankings. OpenAI’s ‘GPT-5.6 Luna’ maintains the top position according to the platform’s ‘Composite Index’. This proprietary metric evaluates models based on human preference, independent benchmarks, real-world traffic, and ecosystem penetration, providing a ranking that reflects actual usage rather than just laboratory performance.