Model Guide
Last updated April 10, 2026
TuneSalon supports 11 models, from lightweight 4B parameter models to powerful 35B models. Smaller models train faster and respond quicker. Larger models produce more nuanced output. Use this guide to find the right model for your use case.
Find Your Model
Step 1: What do you want to build?
All Models
Full comparison of all 11 supported models. VRAM listed is for running the model locally at full precision (FP16) without quantization.
| Model | Capability | Best for | GPU | Local VRAM | GGUF Export |
|---|---|---|---|---|---|
| Qwen3-4B 4B | Lightweight | Multilingual tasks, concise writing, lightweight assistants. | A100 | ~10 GB | ✓ |
| Mistral-7B 7B | Capable | Content writing, general knowledge, customer-facing text. | A100 | ~17 GB | ✓ |
| Qwen3-8B 8B | Capable | Balanced quality and speed. Strong all-rounder for most use cases. | A100 | ~20 GB | ✓ |
| Ministral-8B 8B | Capable | Structured output, data extraction, technical writing. | A100 | ~22 GB | ✓ |
| Qwen3-14B 14B | Advanced | Deep domain knowledge, nuanced writing, multilingual fluency. | A100 | ~36 GB | ✓ |
| Mistral-Small-24B 24B | Advanced | Professional-grade content, complex reasoning, specialist fields. | A100 | ~55 GB | ✓ |
| Qwen3-32B 32B | Most Powerful | Maximum quality for demanding tasks. Legal, medical, research. | H200 Only | ~80 GB | ✓ |
| Qwen3.5-35B 35B | Most Powerful | Qwen3.5 MoE flagship. 35B total, 3B active params per token. Beta (Engine B). | H200 Only | ~90 GB | ✓ |
| Qwen3.5-27B 27B | Advanced | Qwen's latest dense model. Strong reasoning and instruction following. Beta (Engine B). | A100 | ~56 GB | ✓ |
| Gemma-4-26B-A4B 26B (MoE) | Advanced | Google's Gemma 4 MoE. 26B total, 4B active. Efficient and capable. Beta (Engine B). | A100 | ~52 GB | ✓ |
| Gemma-4-31B 31B | Most Powerful | Google's Gemma 4 dense flagship. Top-tier instruction following. Beta (Engine B). | H200 Only | ~62 GB | ✓ |
Multilingual tasks, concise writing, lightweight assistants.
Content writing, general knowledge, customer-facing text.
Balanced quality and speed. Strong all-rounder for most use cases.
Structured output, data extraction, technical writing.
Deep domain knowledge, nuanced writing, multilingual fluency.
Professional-grade content, complex reasoning, specialist fields.
Maximum quality for demanding tasks. Legal, medical, research.
Qwen3.5 MoE flagship. 35B total, 3B active params per token. Beta (Engine B).
Qwen's latest dense model. Strong reasoning and instruction following. Beta (Engine B).
Google's Gemma 4 MoE. 26B total, 4B active. Efficient and capable. Beta (Engine B).
Google's Gemma 4 dense flagship. Top-tier instruction following. Beta (Engine B).