MTEB(Multilingual, v2): MMTEB Multilingual v2
Multilingual text-embedding benchmark (MMTEB) covering 131 tasks in more than 250 languages, including bitext mining, classification, clustering, retrieval, reranking, instruction reranking and semantic similarity. The score is the mean of per-task main scores. Higher is better.
Models ranked
8
tracked in MTEB(Multilingual, v2)
Score band
74.3 – 51.9
best → lowest tracked
Snapshot trend
—
MTEB(Multilingual, v2) · need ≥2 snapshots
Published snapshot
MTEB(Multilingual, v2) · observed Oct 17, 2026 · 8 model variants.
Notes: MTEB(Multilingual, v2) Mean (Task) 74.27 over 131 tasks; Mean (TaskType) 64.20. Computed with the leaderboard's own code (mteb 2.26.6 Benchmark summary table) from embeddings-benchmark/results @ 4bb41dd (2026-10-17). Source: MTEB (embeddings-benchmark/results, CC0). Retrieved 2026-10-17. Microsoft's harrier-oss-v1 card reports 74.3 (1 dp), consistent with this value.
Notes: MTEB(Multilingual, v2) Mean (Task) 70.58 as published on the Qwen3-Embedding model card (standing rule: the model's own card wins). The leaderboard aggregation of embeddings-benchmark/results @ 4bb41dd gives 70.58 on 2026-10-17 (rounding only). Source: Qwen3-Embedding card; MTEB (embeddings-benchmark/results, CC0). Retrieved 2026-10-17.
Notes: MTEB(Multilingual, v2) Mean (Task) 69.01 over 131 tasks; Mean (TaskType) 59.00. Computed with the leaderboard's own code (mteb 2.26.6 Benchmark summary table) from embeddings-benchmark/results @ 4bb41dd (2026-10-17). Source: MTEB (embeddings-benchmark/results, CC0). Retrieved 2026-10-17. Microsoft's harrier-oss-v1 card reports 69.0 (1 dp), consistent with this value.
Notes: MTEB(Multilingual, v2) Mean (Task) 68.37 over 131 tasks; Mean (TaskType) 59.59. Computed with the leaderboard's own code (mteb 2.26.6 Benchmark summary table) from embeddings-benchmark/results @ 4bb41dd (2026-10-17). Source: MTEB (embeddings-benchmark/results, CC0). Retrieved 2026-10-17.
Notes: MTEB(Multilingual, v2) Mean (Task) 66.55 over 131 tasks; Mean (TaskType) 56.90. Computed with the leaderboard's own code (mteb 2.26.6 Benchmark summary table) from embeddings-benchmark/results @ 4bb41dd (2026-10-17). Source: MTEB (embeddings-benchmark/results, CC0). Retrieved 2026-10-17. Microsoft's harrier-oss-v1 card reports 66.5 (1 dp), consistent with this value.
Notes: MTEB(Multilingual, v2) Mean (Task) 64.33 as published on the Qwen3-Embedding model card (standing rule: the model's own card wins). The leaderboard aggregation of embeddings-benchmark/results @ 4bb41dd gives 64.34 on 2026-10-17 (rounding only). Source: Qwen3-Embedding card; MTEB (embeddings-benchmark/results, CC0). Retrieved 2026-10-17.
Notes: MTEB(Multilingual, v2) Mean (Task) 55.96 over 131 tasks; Mean (TaskType) 49.35. Computed with the leaderboard's own code (mteb 2.26.6 Benchmark summary table) from embeddings-benchmark/results @ 4bb41dd (2026-10-17). Source: MTEB (embeddings-benchmark/results, CC0). Retrieved 2026-10-17.
Notes: MTEB(Multilingual, v2) Mean (Task) 51.92 over 131 tasks; Mean (TaskType) 45.91. Computed with the leaderboard's own code (mteb 2.26.6 Benchmark summary table) from embeddings-benchmark/results @ 4bb41dd (2026-10-17). Source: MTEB (embeddings-benchmark/results, CC0). Retrieved 2026-10-17.
Attribution. Source: MTEB (embeddings-benchmark/results, CC0)
Multilingual edition: Enevoldsen et al., MMTEB (arXiv:2502.13595).
How to read this benchmark
This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.
Trust this score when
- There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
- The model list covers the same version family you can actually deploy today.
- Top candidates overlap with your required routing and feature requirements.
Be cautious when
- There is only one benchmark snapshot or the dataset appears stale.
- The benchmark metric direction is opposite of your decision objective.
- The score difference between options is narrow and likely within implementation variance.
Related benchmarks
Last reviewed: Oct 17, 2026