Qwen3.5-27B Opus v2 - Reasoning Model
Base Model: Qwen3.5-27B BF16
Architecture: Qwen3.5 MoE
Purpose: Complex reasoning, chain-of-thought, math, science, analysis
Context: 262K tokens
Recommended Settings
Thinking Mode (Default - Recommended)
Use for: Math, science, logic, complex reasoning, multi-step problems
--temp 0.6 --top-p 0.95 --top-k 20 --min-p 0 --reasoning onNon-Thinking Mode
Use for: Creative writing, casual chat, fast responses
--temp 0.7 --top-p 0.8 --top-k 20 --min-p 0Quantization Guide
| Quant | Size | Quality | Use Case |
|---|
| Q4_K_M | 16GB | ⭐⭐⭐⭐⭐ | Daily use (recommended) |
| Q5_K_M | 18GB | ⭐⭐⭐⭐⭐ | High quality tasks |
| Q6_K | 22GB | ⭐⭐⭐⭐⭐⭐ | Near-lossless |
| Q8_0 | 28GB | ⭐⭐⭐⭐⭐⭐⭐ | Maximum quality |
| BF16 | 54GB | ⭐⭐⭐⭐⭐⭐⭐ | Original quality |
Quick Start
llama-server -m Qwen3.5-27B-Opus-v2-Q4_K_M.gguf --ctx-size 262144 --temp 0.6 --top-p 0.95 --top-k 20 --reasoning onVision Support
Use with mmproj for multimodal capabilities:
--mmproj mmproj-Qwen3.5-27B-Opus-v2-f16.ggufModel Variants
- Opus v2 (this repo) - Standard 262K context
- Opus v2 YaRN 1M - Extended 1M context (separate repo)
