Solstice-AI/DeepSeek-V4-Flash-Vision-UNCENSORED-mlx-oQ4e-DSpark

DeepSeek-V4-Flash-Vision-UNCENSORED (MLX oQ4e Suite)

Official Solstice-AI Release • Optimized 4-Bit MLX • Lossless BF16 Vision Tower • Bundled MLX DSpark Drafter • Apple Silicon Native

---

Model Overview

Solstice-AI/DeepSeek-V4-Flash-Vision-UNCENSORED-mlx-oQ4e-DSpark provides the official consumer-friendly 4-bit MLX release of DeepSeek-V4-Flash-Vision-UNCENSORED. Optimized specifically for Apple Silicon (M-series MacBooks and Mac Studios with 64GB–96GB unified memory), keeping the 32-layer vision tower in pristine BF16 and bundling pre-aligned MLX DSpark speculative drafters.

Key Specifications

Attribute	Specification
Base Model	orcarouter/DeepSeek-V4-Flash-Vision-Uncensored
Total Parameters	305B (256 routed MoE experts, ~18B active per token)
Context Window	1,048,576 tokens (1M YaRN native context)
Multimodal Vision	32-layer Vision Transformer (ViT) in native BF16 (Shard 1 aligner weights preserved)
Quantization Format	30-shard SafeTensors MLX oQ4e (4-bit MoE, BF16 attention & projections)
Speculative Drafter	Bundled native MLX DSpark speculative drafter suite in speculative/

---

Benchmark Highlights

• Terminal-Bench 2.1: 83.9% (Agentic CLI execution)
• SWE-bench Verified: 65.8% (Real-world software engineering)
• LiveCodeBench v6: 84.2% (Algorithmic problem solving)
• MATH-500: 94.6%
• DocVQA / ChartQA: 92.3% (Complex visual reasoning & document grounding)

---

Attribution & Acknowledgments

• Original Architecture: DeepSeek AI
• Uncensored Alignment: orcarouter
• Quantization & DSpark Packaging: Solstice-AI