Blackfrost-AI/Qwythos-9B-V2-EMERGED-BF16
Blackfrost-AI/Qwythos-9B-V2-EMERGED-BF16 is a 9.5 billion parameter vision-language model developed by Blackfrost AI, based on empero-ai's Qwythos-9B. This model is a pure-reasoning and agentic distillation, trained exclusively on "The Void" corpus to sharpen its reasoning and tool-use capabilities without vertical bias. It supports multimodal inputs including text, images, and video, making it suitable for complex reasoning tasks requiring diverse input modalities.
Loading preview...
Overview of Qwythos-9B-V2-EMERGED-BF16
Qwythos-9B-V2-EMERGED-BF16 is a 9.5 billion parameter vision-language model developed by Blackfrost AI, built upon the empero-ai/Qwythos-9B-Claude-Mythos-5-1M base. It features a qwen3_5 architecture and operates with BF16 precision. A key differentiator is its training on The Void v4, Blackfrost's proprietary corpus of ~5,032 multi-turn distilled reasoning and agentic (ReAct-style tool-use) trajectories. This specialized training focuses on enhancing the model's reasoning core and agentic behavior without introducing domain-specific biases.
Key Capabilities
- Pure Reasoning & Agentic Behavior: Sharpened for complex reasoning and tool-use through distillation on "The Void" corpus, which contains no marketing or domain-specific data.
- Multimodal Vision-Language Model: Inherits a full vision tower from its base, enabling it to process and understand images and video alongside text inputs. (Note: Blackfrost's distillation was text-only; vision capabilities track the base model).
- High Context Length: Supports a native context length of 1,048,576 (1M) tokens, inherited from the base model.
- Designed for "Thinking Enabled" Generation: Optimized to surface distilled reasoning traces, beneficial for tasks requiring transparent thought processes.
When to Use This Model
This model is ideal for use cases requiring strong, unbiased reasoning and agentic capabilities. Its multimodal nature makes it suitable for applications that need to interpret both textual and visual information. Developers seeking a model with a sharpened reasoning core, free from vertical bias, for tasks like complex problem-solving, logical deduction, or tool-use integration, will find Qwythos-9B-V2-EMERGED-BF16 particularly effective. It is part of an A/B design, representing the pure reasoning build, contrasting with versions layered with domain-specific corpora.