Blackfrost-AI/PINQWEN-3.5-9B-1M-BF16

VISIONConcurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 14, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

PINQWEN-3.5-9B-1M-BF16 is a 9 billion parameter reasoning model from Blackfrost AI, built on the Qwen 3.5 9B architecture. It features an exceptional 1,000,000-token context window and a native chain-of-thought reasoning process. This uncensored, multimodal model excels at general reasoning, coding assistance, and technical Q&A, supporting both text and image inputs.

Loading preview...

PINQWEN-3.5-9B-1M: A Compact, Long-Context Reasoning Model

PINQWEN-3.5-9B-1M is Blackfrost AI's 9 billion parameter model, distilled using their multi-teacher reasoning-distillation method, The Void, on the Qwen 3.5 9B architecture. This model is designed to reason transparently using a native <think> block before generating answers, and it operates with an impressive 1,000,000-token context window, making it suitable for handling extensive inputs.

Key Capabilities

  • 1,000,000-token context: Processes entire codebases, books, or long chat logs efficiently, leveraging a YaRN-extended window on a gated-linear-attention hybrid backbone.
  • Uncensored: Provides direct answers without the reflexive refusals common in over-aligned models, allowing users to define its operational scope.
  • Transparent Reasoning: Employs an inspectable <think>...<think> chain-of-thought, distilled from frontier teachers, for clear and readable reasoning.
  • Coding and Technical Proficiency: Tuned for software engineering tasks, step-by-step problem-solving, and precise technical explanations.
  • Multimodal Vision: Includes a full vision encoder, enabling it to process both text and image inputs.
  • Versatile Deployment: Available in BF16 (reference precision), NVFP4 (for NVIDIA Blackwell), and GGUF (for llama.cpp / local inference with a full quantization ladder and MTP speculative-decode head).

Intended Use Cases

  • General reasoning and problem-solving.
  • Coding assistance and technical Q&A.
  • Understanding ultra-long documents and entire codebases.
  • Multimodal applications involving both image and text inputs.