# MTEB(kor, v2) · Korean

> Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/mteb%3AMTEB(kor%2C%20v2). When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/mteb%3AMTEB(kor%2C%20v2).

Korean text embedding quality across classification, clustering, pair classification (NLI), reranking, retrieval, and semantic similarity. Expanded successor to MTEB(kor, v1): grows the suite from 6 to 20 tasks, adding clustering (SIB200, KLUE-MRC domain/category), pair classification/NLI (KLUE-NLI, KorNLI, PawsX), additional retrieval (LawIRKo, SQuAD-Ko, AutoRAG, PublicHealthQA, Belebele, MultiLongDoc, MrTidy) and STS17, for broader and more robust coverage of Korean embedding capabilities.

Canonical URL: https://benchize.com/benchmarks/mteb%3AMTEB(kor%2C%20v2)
Last updated: Wed, 21 Oct 2026 18:52:20 GMT

Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations.

Korean text embedding quality across classification, clustering, pair classification (NLI), reranking, retrieval, and semantic similarity. Expanded successor to MTEB(kor, v1): grows the suite from 6 to 20 tasks, adding clustering (SIB200, KLUE-MRC domain/category), pair classification/NLI (KLUE-NLI, KorNLI, PawsX), additional retrieval (LawIRKo, SQuAD-Ko, AutoRAG, PublicHealthQA, Belebele, MultiLongDoc, MrTidy) and STS17, for broader and more robust coverage of Korean embedding capabilities.

Embeddings · Models tested: 1 · Score publishers: MTEB

## Model results

| # | Model | Score |  |
| --- | --- | --- | --- |
| 1 | [Qwen3 Embedding](https://benchize.com/models/qwen3-embedding.md) | 0.78source score |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `mean_task` \| 0.7773353333333334 \| source_score \| source_published_aggregate \| Higher is better \| \| `mean_task_type` \| 0.7544947901234568 \| source_score \| source_published_aggregate \| Higher is better \| ### Configuration zeroShotPct 85 ### Reported statistics experiments Not reported rank 1 scoresByTask {"AutoRAGRetrieval":0.82765,"BelebeleRetrieval":0.9827066666666667,"KLUE-NLI":0.781156,"KLUE-STS":0.8913,"KLUE-TC":0.690967,"KlueMrcDomainClustering":0.753214,"KlueYnatMrcCategoryClustering":0.811067,"Ko-StrategyQA":0.83633,"KorNLI":0.876054,"KorSTS":0.900656,"LawIRKo":0.81712,"MIRACLReranking":0.67406,"MIRACLRetrieval":0.67829,"MrTidyRetrieval":0.61873,"MultiLongDocRetrieval":0.504565,"PawsXPairClassification":0.660622,"PublicHealthQA":0.87211,"SIB200ClusteringS2S":0.562952,"SQuADKorV1Retrieval":0.90628,"STS17":0.900877} scoresByTaskType {"Classification":0.690967,"Clustering":0.7090776666666668,"PairClassification":0.7726106666666667,"Reranking":0.67406,"Retrieval":0.7826424074074075,"STS":0.897611} trainedOnTasks ["MIRACLReranking","MIRACLRetrieval","MrTidyRetrieval"] ### Source [Original source](https://mteb-leaderboard-backend.hf.space/v1/benchmarks/MTEB(kor%2C%20v2)/scores) Source MTEB Original model label Qwen/Qwen3-Embedding-8B Observed 2026-10-21T16:40:34.371Z Evidence SHA-256 `e1954767ff1c78bb6e47d1883fb79b56cb2a2d19d2749611fb5bb71bd86c2169` Source locator `/rows/0` |  |  |  |
| 2 | [Qwen3 Embedding](https://benchize.com/models/qwen3-embedding.md) | 0.77source score |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `mean_task` \| 0.7694064083333334 \| source_score \| source_published_aggregate \| Higher is better \| \| `mean_task_type` \| 0.7467482067901235 \| source_score \| source_published_aggregate \| Higher is better \| ### Configuration zeroShotPct 85 ### Reported statistics experiments Not reported rank 2 scoresByTask {"AutoRAGRetrieval":0.84314,"BelebeleRetrieval":0.9483366666666667,"KLUE-NLI":0.783378,"KLUE-STS":0.892954,"KLUE-TC":0.673877,"KlueMrcDomainClustering":0.870177,"KlueYnatMrcCategoryClustering":0.745475,"Ko-StrategyQA":0.82697,"KorNLI":0.867713,"KorSTS":0.899716,"LawIRKo":0.7769,"MIRACLReranking":0.67252,"MIRACLRetrieval":0.68028,"MrTidyRetrieval":0.6076,"MultiLongDocRetrieval":0.502245,"PawsXPairClassification":0.6073955,"PublicHealthQA":0.86926,"SIB200ClusteringS2S":0.51149,"SQuADKorV1Retrieval":0.90445,"STS17":0.904251} scoresByTaskType {"Classification":0.673877,"Clustering":0.7090473333333334,"PairClassification":0.7528288333333334,"Reranking":0.67252,"Retrieval":0.7732424074074075,"STS":0.8989736666666667} trainedOnTasks ["MIRACLReranking","MIRACLRetrieval","MrTidyRetrieval"] ### Source [Original source](https://mteb-leaderboard-backend.hf.space/v1/benchmarks/MTEB(kor%2C%20v2)/scores) Source MTEB Original model label Qwen/Qwen3-Embedding-4B Observed 2026-10-21T16:40:34.371Z Evidence SHA-256 `e1954767ff1c78bb6e47d1883fb79b56cb2a2d19d2749611fb5bb71bd86c2169` Source locator `/rows/1` |  |  |  |

---
Cite as: MTEB(kor, v2) · Korean — Benchize, https://benchize.com/benchmarks/mteb%3AMTEB(kor%2C%20v2) (retrieved 2026-10-21)