Ministral-3-Reasoning-2512-AIO-GGUF

Ministral-3-Reasoning-2512-AIO-GGUF

The Ministral 3 Reasoning models (3B, 8B, and 14B variants from mistralai) are post-trained vision-language models specialized for advanced reasoning tasks like math, coding, and STEM applications, featuring a core language model (3.4B, 8.4B, or 13.5B parameters) paired with a 0.4B vision encoder for multimodal image analysis, supporting a 256k context window, multilingual capabilities, and edge deployment on hardware as low as 24GB VRAM/RAM when quantized (BF16 precision). Optimized with a recommended temperature of 0.7 and top_p=0.95 for reasoning, they use a distinctive chat template encouraging structured [THINK] inner monologue drafts in Markdown/LaTeX before final responses, enabling step-by-step problem-solving while maintaining strong performance in benchmarks like AIME25 (0.721 for 3B) and GPQA Diamond. Ideal for resource-efficient local inference via vLLM or Transformers, these Apache 2.0-licensed models excel in agentic workflows, function calling, and complex multimodal reasoning under constrained environments.

Ministral-3-14B-Reasoning-2512 [GGUF]

Ministral-3-8B-Reasoning-2512 [GGUF]

Ministral-3-3B-Reasoning-2512 [GGUF]

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):