Ternlight
Ranked 2 of 8 embedding models by recommenders
Tiny embedding model that runs in the browser for on-device semantic search without API calls. Caveat: heavy CPU and memory use, and failures on some browsers and phones.
Tiny embedding model that runs in the browser for on-device semantic search without API calls. Caveat: heavy CPU and memory use, and failures on some browsers and phones.
20 people gave an opinion
- recommend it
- 17
- criticize it
- 3
- approval share
- 85%
Recommended in 1 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: not recommended
- February 2026: not recommended
- March 2026: not recommended
- April 2026: not recommended
- May 2026: not recommended
- June 2026: not recommended
- July 2026: recommended
- August 2026: not recommended
- September 2026: not recommended
- October 2026: not recommended
- November 2026: not recommended (month in progress)
Why it doesn’t carry the badge
- 17 people recommend itNeeds 10 or more, met.
- 5.7 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 1 of 13 monthsNeeds 4 or more, not met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Fast, instant on-device search
“Running it on-device means search-as-you-type semantic search is performant with no API dependencies.”
Comment from 6 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” “allowing us to search in the 2 sources instantly”
Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” “Inference is nice and quick after that.”
Comment from 6 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” - Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)”
Praised as impressive, promising project
“Interesting project. Happy to see someone who shares an interest in tiny vector embeddings models.”
Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” “could be the missing piece for something I wanted to build”
Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” “That's really impressive, congratulations.”
Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” Strong fidelity to MiniLM teacher
“0.84 Spearman fidelity to the MiniLM teacher at ternary precision is a striking result”
Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” “the loss is not linear, looks like you give up less than 1/3 of information”
Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)”
What people warn about
Each mark is one comment
Heavy CPU and memory use
“it's a bit startling to hear my fans go crazy when opening a webpage”
Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” “This could be used to distribute malware and also or hog excessive browser memory.”
Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” Fails on some browsers and phones
“Does not work on iPhone 11 Pro Max and iPhone 16 Pro.”
Comment from 7 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)” “Also does not work in Firefox 152.0.6 (aarch64). Due to WebGPU issue.”
Comment from 17 Jul 2026 in “Ternlight – 7 MB embedding model that runs in browser (WASM)”
Other embedding models people recommend
The four most recommended other embedding models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| EmbeddingGemma | 27 | 4 | 87% | 3 | — | — |
| TernlightThis page | 17 | 3 | 85% | 1 | — | — |
| CLIP by OpenAI | 10 | 5 | 67% | 6 | — | Approved |
| Gemini Embedding 2 | 9 | 3 | 75% | 3 | — | — |
| Qwen3-Embedding | 8 | 1 | 89% | 4 | — | — |
What people said
18 of the 22 comments that mention it, newest first. Each links to the original comment.
“Also does not work in Firefox 152.0.6 (aarch64). Due to WebGPU issue.”
“could be the missing piece for something I wanted to build”
“Demo works quite strangely. Similarity search failed.”
“the loss is not linear, looks like you give up less than 1/3 of information”
“allowing us to search in the 2 sources instantly”
“Prime example of wasm supremacy over JavaScript.”
“Huge kudos for sharing everything including your training code.”
“But the claimed performance seems way off, I get only 35 emb/sec in firefox on a i5-4570 rather than 400/sec.”
“cheap and fast search in a product base”
“0.84 Spearman fidelity to the MiniLM teacher at ternary precision is a striking result”
“Interesting project. Happy to see someone who shares an interest in tiny vector embeddings models.”
“Does not work on iPhone 11 Pro Max and iPhone 16 Pro.”
“That's really impressive, congratulations.”
“This could be used to distribute malware and also or hog excessive browser memory.”
“it's good for basic math and coding”
“it's a bit startling to hear my fans go crazy when opening a webpage”
“Inference is nice and quick after that.”
“Running it on-device means search-as-you-type semantic search is performant with no API dependencies.”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.