# all-MiniLM-L6-v2

6 Hacker News commenters recommended all-MiniLM-L6-v2 and 3 criticized it, November 2025 to November 2026. It was recommended in 6 of 13 months.

**Verdict:** A small, free, open embedding model that runs on-device and gives good semantic matching for its size. Critics say it is old and behind newer open-weight models.

## Why people recommend it

- Small enough to run on-device (2 comments): "embedding models we had for years like minlm (80mb) are so small, and I really wonder why not more on device searches use something like it" ([source](https://news.ycombinator.com/item?id=46172711)) "One thing that's still compelling about all-Mini is that it's feasible to use it client-side." ([source](https://news.ycombinator.com/item?id=46082499))
- Good semantic matching for its size (2 comments): "Works well enough for semantic matching against ~50k transaction descriptions." ([source](https://news.ycombinator.com/item?id=46631165)) "only 1.3 points behind bge-large at 15x smaller" ([source](https://news.ycombinator.com/item?id=49833068))

## What people warn about

- Outdated compared with newer models (3 comments): "it's old and does not implement the newest advances in architectures and data training pipelines" ([source](https://news.ycombinator.com/item?id=46081800)) "This is so old and terribly behind the new models" ([source](https://news.ycombinator.com/item?id=46082011)) "there's much better open weights models out there" ([source](https://news.ycombinator.com/item?id=46082012))

## Alternatives hackers recommend

- [EmbeddingGemma](https://hackerapproved.com/p/embeddinggemma-2.md): 27 recommend
- [Ternlight](https://hackerapproved.com/p/ternlight.md): 17 recommend
- [CLIP by OpenAI](https://hackerapproved.com/p/clip.md): 10 recommend
- [Gemini Embedding 2](https://hackerapproved.com/p/gemini-embedding-2.md): 9 recommend
- [Qwen3-Embedding](https://hackerapproved.com/p/qwen3-embedding-0-6b.md): 8 recommend

## Cite as

Hacker Approved, "all-MiniLM-L6-v2", https://hackerapproved.com/p/all-minilm-l6-v2/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).