S1-mini
By superwhisper · United States
Updated 2026-08-31
Overview
S1-mini is superwhisper's 0.6B speech-recognition model built on Qwen3-0.6B, offering fully local, ultra-lightweight transcription at roughly 0.3GB VRAM in Q4.
When to pick this model
- On-device transcription without cloud dependency
- Offline dictation on laptops or edge devices
- Low-power ASR for mobile or embedded deployments
- Privacy-sensitive audio transcription
VRAM requirements by quantization
| Quantization | VRAM required |
|---|---|
| Q4_K_M (recommended) | 0.3 GB |
| Q5_K_M | 0.4 GB |
| Q8_0 | 0.6 GB |
| FP16 (no quantization) | 1.2 GB |
VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.
In practice, S1-mini fits an 8 GB consumer card at Q4_K_M (0.3 GB). Stepping up to Q8_0 nearly doubles the footprint to 0.6 GB, and unquantized FP16 weights take 1.2 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.
Without a GPU, S1-mini needs roughly 0.8 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 110 tokens/sec on entry-level GPUs, on the order of 170 tokens/sec on a mid-range card, and up to 220 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.
What hardware do you need
The table below matches S1-mini to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.
| GPU memory | Example cards | Best fit for S1-mini |
|---|---|---|
| 8 GB | RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB | FP16 (1.2 GB used) |
| 12 GB | RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop | FP16 (1.2 GB used) |
| 16 GB | RTX 5080, RTX 4080 Super, Radeon RX 9070 XT | FP16 (1.2 GB used) |
| 24 GB | RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop | FP16 (1.2 GB used) |
| 32 GB | RTX 5090 | FP16 (1.2 GB used) |
Which hardware should you buy to run S1-mini?
To run S1-mini locally at Q4, you need ~0.3 GB of VRAM. The best value for this today is a RTX 5060 Ti 16GB (ASUS Dual OC) (16 GB VRAM, best $/GB).
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
Strengths
- Extremely lightweight — 0.6B, ~0.3GB VRAM at Q4
- Fully local transcription, no cloud round-trip
- Very fast inference, suitable for edge and mobile
- Built on the Qwen3-0.6B base
Limitations
- Primarily English-focused
- "Other" license — check usage terms before deployment
- No Ollama tag — install from Hugging Face
Typical workloads
In our catalog grid, S1-mini is filed under Local Transcription, Edge ASR, Offline Dictation — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: audio understanding.
The 40k-token context window covers long chats and mid-sized documents, though very large retrieval workloads will need chunking. It ships under the Autre (open weights) license — commercial use is generally possible but read the specific terms before embedding it in a product.
Architecture & training
Architecture: Dense transformer · 0.6B parameters · 40K context · based on Qwen3-0.6B
Training: Speech-recognition (ASR) model released by superwhisper, derived from Qwen3-0.6B. 'Other' license (open weights).
A tiny, fast, fully local ASR model — ideal for offline dictation, not a multilingual transcription solution.
Quick start
# HuggingFace : superwhisper/s1-miniOr use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.
Similar models worth comparing
Frequently asked questions
How much VRAM does S1-mini need?
At the recommended Q4_K_M quantization, S1-mini needs about 0.3 GB of VRAM. Q8_0 takes 0.6 GB, and unquantized FP16 weights take 1.2 GB.
Can S1-mini run without a GPU?
Yes — with roughly 0.8 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.
What context window does S1-mini support?
S1-mini supports a 40k-token context window (40,960 tokens).
Can I use S1-mini commercially?
S1-mini ships under the Autre (open weights) license. Commercial use is generally permitted subject to its terms — review the license text before shipping a product.
How fast is S1-mini on consumer hardware?
Our compatibility engine estimates on the order of 170 tokens/sec on a mid-range GPU and up to 220 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.
Which quantization of S1-mini should I download first?
Start with Q4_K_M (0.3 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It already fits an 8 GB card at FP16.