LateOn measured on 6 of 261 Embeddings benchmarks6
BenchmarkByScoreRank
No benchmark matches
self-host onlyprice
57.2BEIR
Apr 2026released
0K tokensMax input
enLanguages
Alternatives?
ModelScoreCostPriceWhat changes
BetterQwen 3 Embedding 8BunstatedOpen+10%—$0.01Languages: not verified+1
OpenNV Embed v2unstated+10%——Memory (4-bit): 0.1 GB → 4.7 GB
BetterBge En IclunstatedOpen+9%—$0.01Languages: not verified+1BetterQwen 3 Embedding 4BunstatedOpen+8%—$0.02Languages: not verified+1BetterBge Multilingual Gemma2unstatedOpen+4%—$0.11Languages: not verified+1ModelScoreCostPriceWhat changes
Nothing here yet.
ModelScoreCostPriceWhat changes
Qwen 3 Embedding 8BunstatedOpen+10%—$0.01Languages: not verified+1Bge En IclunstatedOpen+9%—$0.01Languages: not verified+1Qwen 3 Embedding 4BunstatedOpen+8%—$0.02Languages: not verified+1Bge Multilingual Gemma2unstatedOpen+4%—$0.11Languages: not verified+1VVoyage 3 M Expunstated+19%——Languages: not verifiedModelScoreCostPriceWhat changes
NV Embed v2unstatedOpen+10%——Memory (4-bit): 0.1 GB → 4.7 GBQwen 3 Embedding 8BunstatedOpen+10%—$0.01Languages: not verified+1Bge En IclunstatedOpen+9%—$0.01Languages: not verified+1yLENS D8000unstatedOpen+8%——Languages: not verified+1iInf Retriever v1unstatedOpen+8%——Memory (4-bit): 0.1 GB → 4.2 GBModelScoreCostPriceWhat changes
lmLateOnOpensame——Memory (4-bit): 0.1 GB → 0.2 GBlmDenseOnOpensame——Memory (4-bit): 0.1 GB → 0.2 GBlDenseOnOpen-2%——lColBERT ZeroOpen-3%——lGTE ModernColBERT v1Open-4%——lColBERT Zero SupervisedOpen-9%——lColBERT Zero UnsupervisedOpen-10%——lDenseOn UnsupervisedOpen-14%——All lightonai models →Run it yourself
estimateFits on✓ 8 GB (phone, laptop) · full quality✓ 16 GB · full quality✓ 24 GB GPU · full quality✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
LicenceApache-2.0free for commercial use
Benchmarks
in Embeddings & RetrievalBEIR · Retrieval · English57.2#26
FAQ
How much does LateOn cost?›
There is no per-unit API price: self-host only.
Is LateOn free to use?›
There is no free API. The weights are open, so you can run it yourself.
How good is LateOn, and what is it best at?›
#26 of 194 in Embeddings.
Is LateOn better than F2LLM v2 8B?›
On BEIR · Mean over task types · English, LateOn scores 57.2 vs 57.5 — about the same.
When was LateOn released? Is there a newer version?›
Released 2026-04-21. It is the current version in its line.
Is LateOn open source?›
Yes, the weights are open (Apache-2.0 licence).
Can I run LateOn locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
What is LateOn?›
LateOn is an open-weights embeddings & retrieval model from lightonai, released Apr 2026.
What are the best alternatives to LateOn?›
Qwen 3 Embedding 8B (better), NV Embed v2 (open weights).