all-MiniLM-L6-v2
all-MiniLM-L6-v2 is a 22M-parameter sentence-transformers embedder producing 384-dim vectors with a 256-token context, distilled from a larger Microsoft MiniLM teacher and fine-tuned on 1B+ sentence pairs across 32 datasets. It is the canonical default embedder for browser-side RAG, Transformers.js demos, and Chroma's quick-start — the most-downloaded sentence-transformers model on HuggingFace by a wide margin.
The 'just embed something' default. all-MiniLM-L6-v2 is what every quick-start tutorial uses for a reason: it fits in 100MB, it runs in any runtime including the browser, and the Apache license has zero friction. For production RAG with quality demands, upgrade to nomic-embed-text-v1.5 or mxbai-embed-large-v1. But for prototypes, in-browser demos, and any deployment where storage cost dominates, this is still the right answer in 2026.
Overview
all-MiniLM-L6-v2 is a 22M-parameter sentence-transformers embedder producing 384-dim vectors with a 256-token context, distilled from a larger Microsoft MiniLM teacher and fine-tuned on 1B+ sentence pairs across 32 datasets. It is the canonical default embedder for browser-side RAG, Transformers.js demos, and Chroma's quick-start — the most-downloaded sentence-transformers model on HuggingFace by a wide margin.
Strengths
- 22M params — sub-100MB footprint, runs in a browser tab or on a Raspberry Pi Zero
- 384-dim output is 1/3 the storage cost of 1024-dim BGE/mxbai vectors
- Apache-2.0 with no acceptable-use restrictions
- Ubiquitous: Chroma, LangChain, LlamaIndex, Transformers.js, fastembed all ship it as default
- ONNX, CoreML, and Transformers.js artifacts maintained by Xenova and others
Weaknesses
- MTEB English score (~56.3) trails BGE/mxbai-large by ~8 points — quality ceiling is real
- 256-token context is the shortest in the embedder catalog — chunking is mandatory
- English-only; for multilingual use paraphrase-multilingual-MiniLM-L12-v2
- Symmetric similarity only — no query/document distinction (use multi-qa-MiniLM if you need that)
Quantization variants
Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.
| Quantization | File size | VRAM required |
|---|---|---|
| Q4_K_M | 0.0 GB | 1 GB |
Get the model
HuggingFace
Original weights
Source repository — direct quantization required.
Hardware that runs this
Cards with enough VRAM for at least one quantization of all-MiniLM-L6-v2.
Models worth comparing
Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.
Frequently asked
What's the minimum VRAM to run all-MiniLM-L6-v2?
Can I use all-MiniLM-L6-v2 commercially?
What's the context length of all-MiniLM-L6-v2?
Source: huggingface.co/sentence-transformers/all-MiniLM-L6-v2
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.
Related — keep moving
Verify all-MiniLM-L6-v2 runs on your specific hardware before committing money.