MTEB-BR
Embedding models encode Brazilian Portuguese text for classification, clustering, similarity, retrieval, and reranking tasks. The headline score averages the primary task metrics across 22 native Portuguese tasks.
The suite excludes translated corpora. It differs from the general MTEB task selection. Task-level metrics and bootstrap intervals accompany the aggregate.
- Released
- July 2026
- Kind
- Reading in many languages
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- Portuguese
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- Based on
- MTEB
- Last model added
- March 2026
- Scores refreshed
- November 20th, 2026
In the authors’ words
Text embeddings for Portuguese have no dedicated benchmark: evaluation rests on translated corpora such as English MS MARCO or on thin multilingual coverage, with native tasks scattered and unconsolidated. We introduce MTEB-BR, a benchmark of 22 native Brazilian-Portuguese tasks across seven categories (classification, multilabel classification, pair classification, semantic textual similarity, clustering, retrieval, and reranking), admitting only data created or found in Portuguese and excluding translations by construction. We evaluate 93 models spanning 23M to 27B parameters: 73 open-weight and 20 closed commercial APIs. Alongside the leaderboard we report a statistical layer for every headline comparison: per-task bootstrap confidence intervals, paired-bootstrap significance, a task- and instance-level discrimination analysis (how sharply each task separates models) adapted from Item Response Theory, and a cross-leaderboard correlation. Three findings stand out. The benchmark cleanly separates about a dozen tiers of models, though the top six are statistically too close to order. An openly licensed, self-hostable model reaches that leading tier, so strong Portuguese embedding quality does not require a commercial API. And a model's rank on the global multilingual leaderboard predicts its Portuguese rank only moderately (Spearman rho = 0.75 over 55 shared models; one model ranks 3rd there and 49th here), so a native benchmark measures something the multilingual boards do not. We release every task, our code, and a public leaderboard, so practitioners can choose Portuguese embedding models on native evidence.
gemini-embedding-001 tops the leaderboard with 0.68 Qwen3-Embedding-8B leads with 0.67 among models released by March 2026
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Qwen3-Embedding-8B, at 0.67. This view shows only models we track, each at its best.
The newest model with a score here came out in March 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 0.68 gemini-embedding-001
- 0.67
Qwen3-Embedding-8B
- 0.67 KaLM-Embedding-Gemma3-12B-2511
- 0.67 voyage-context-4
- 0.67 Octen-Embedding-8B
- 0.66
Qwen3-Embedding-4B
- 0.66 voyage-context-3
- 0.66 voyage-3-large
- 0.65 voyage-4-large
- 0.65 SFR-Embedding-Mistral
- 0.65 BidirLM-1.7B-Embedding
- 0.65 BOOM_4B_v1
- 0.65 embeddinggemma-300m
- 0.65 codestral-embed
- 0.65 Linq-Embed-Mistral
- 0.64 text-embedding-3-large
- 0.64 jina-embeddings-v5-text-small
- 0.64 voyage-4
- 0.64 multilingual-e5-large-instruct
- 0.64 SFR-Embedding-2_R
- 0.64 gte-Qwen2-7B-instruct
- 0.64 harrier-oss-v1-27b
- 0.64 BidirLM-1B-Embedding
- 0.64 F2LLM-v2-8B
- 0.64 voyage-3.5
- 0.63 harrier-oss-v1-0.6b
- 0.63 F2LLM-v2-14B
- 0.63
Gemini Embedding 2
- 0.63 F2LLM-v2-4B
- 0.63 PwC-Embedding_expr
- 0.63 voyage-finance-2
- 0.63 Octen-Embedding-0.6B
- 0.62 PIXIE-Rune-v1.0
- 0.62
Qwen3-Embedding-0.6B
- 0.62 gte-Qwen2-1.5B-instruct
- 0.62 voyage-law-2
- 0.62 snowflake-arctic-embed-l-v2.0
- 0.62 voyage-3.5-lite
- 0.62 text-embedding-3-small
- 0.62 bge-m3
- 0.62 voyage-4-lite
- 0.61 F2LLM-v2-1.7B
- 0.61 BidirLM-0.6B-Embedding
- 0.61 voyage-3
- 0.61 jua-4B-mixed
- 0.61 F2LLM-0.6B
- 0.60 harrier-oss-v1-270m
- 0.60 F2LLM-v2-0.6B
- 0.60 mistral-embed
- 0.59 titan-embed-text-v2
- 0.59 multilingual-e5-large
- 0.59 granite-embedding-311m-multilingual-r2
- 0.59 jua-4B-legal-only
- 0.59 voyage-3-lite
- 0.58 F2LLM-v2-330M
- 0.58 multilingual-e5-base
- 0.57 text-embedding-005
- 0.57 serafim-100m-portuguese-pt-sentence-encoder-ir
- 0.57 serafim-335m-portuguese-pt-sentence-encoder-ir
- 0.56 multilingual-e5-small
- 0.56 granite-embedding-107m-multilingual
- 0.56 granite-embedding-97m-multilingual-r2
- 0.56 serafim-900m-portuguese-pt-sentence-encoder-ir
- 0.55 serafim-335m-portuguese-pt-sentence-encoder
- 0.54 serafim-900m-portuguese-pt-sentence-encoder
- 0.54 F2LLM-v2-160M
- 0.53 paraphrase-multilingual-mpnet-base-v2
- 0.53 serafim-100m-portuguese-pt-sentence-encoder
- 0.52 Qwen3-Embedding-0.6B-jua-V2
- 0.51 F2LLM-v2-80M
- 0.50 LaBSE
- 0.50 medlink-bi-encoder
- 0.47 e5-small-v2
- 0.47 mxbai-embed-large-v1
- 0.47 bge-small-en-v1.5
- 0.47 bert-large-portuguese-cased
- 0.46 gte-small
- 0.46 bert-base-portuguese-cased
- 0.46 bert-large-portuguese-cased-legal-mlm-sts-v1.0
- 0.45 e5-mistral-7b-instruct
- 0.45 llama-embed-nemotron-8b
- 0.44 Legal-BERTimbau-sts-large
- 0.44 bert-large-portuguese-cased-legal-mlm-mkd-nli-sts-v1
- 0.44 Ivysaur
- 0.43 albertina-900m-portuguese-ptbr-encoder
- 0.43 all-MiniLM-L12-v2
- 0.42 GIST-all-MiniLM-L6-v2
- 0.42 all-mpnet-base-v2
- 0.41 legal-bert-pt-br
- 0.41 all-MiniLM-L6-v2
- 0.41 Legal-BERTimbau-sts-large-ma-v3
- 0.36 bert-large-portuguese-cased-legal-tsdae-gpl-nli-sts-MetaKD-v0
- 0.25 paraphrase-multilingual-MiniLM-L12-v2
- 1 gemini-embedding-001 0.68
- 2
Qwen3-Embedding-8B 0.67
- 3 KaLM-Embedding-Gemma3-12B-2511 0.67
- 4 voyage-context-4 0.67
- 5 Octen-Embedding-8B 0.67
- 6
Qwen3-Embedding-4B 0.66
- 7 voyage-context-3 0.66
- 8 voyage-3-large 0.66
- 9 voyage-4-large 0.65
- 10 SFR-Embedding-Mistral 0.65
- 11 BidirLM-1.7B-Embedding 0.65
- 12 BOOM_4B_v1 0.65
- 13 embeddinggemma-300m 0.65
- 14 codestral-embed 0.65
- 15 Linq-Embed-Mistral 0.65
- 16 text-embedding-3-large 0.64
- 17 jina-embeddings-v5-text-small 0.64
- 18 voyage-4 0.64
- 19 multilingual-e5-large-instruct 0.64
- 20 SFR-Embedding-2_R 0.64
- 21 gte-Qwen2-7B-instruct 0.64
- 22 harrier-oss-v1-27b 0.64
- 23 BidirLM-1B-Embedding 0.64
- 24 F2LLM-v2-8B 0.64
- 25 voyage-3.5 0.64
- 26 harrier-oss-v1-0.6b 0.63
- 27 F2LLM-v2-14B 0.63
- 28
Gemini Embedding 2 0.63
- 29 F2LLM-v2-4B 0.63
- 30 PwC-Embedding_expr 0.63
- 31 voyage-finance-2 0.63
- 32 Octen-Embedding-0.6B 0.63
- 33 PIXIE-Rune-v1.0 0.62
- 34
Qwen3-Embedding-0.6B 0.62
- 35 gte-Qwen2-1.5B-instruct 0.62
- 36 voyage-law-2 0.62
- 37 snowflake-arctic-embed-l-v2.0 0.62
- 38 voyage-3.5-lite 0.62
- 39 text-embedding-3-small 0.62
- 40 bge-m3 0.62
- 41 voyage-4-lite 0.62
- 42 F2LLM-v2-1.7B 0.61
- 43 BidirLM-0.6B-Embedding 0.61
- 44 voyage-3 0.61
- 45 jua-4B-mixed 0.61
- 46 F2LLM-0.6B 0.61
- 47 harrier-oss-v1-270m 0.60
- 48 F2LLM-v2-0.6B 0.60
- 49 mistral-embed 0.60
- 50 titan-embed-text-v2 0.59
- 51 multilingual-e5-large 0.59
- 52 granite-embedding-311m-multilingual-r2 0.59
- 53 jua-4B-legal-only 0.59
- 54 voyage-3-lite 0.59
- 55 F2LLM-v2-330M 0.58
- 56 multilingual-e5-base 0.58
- 57 text-embedding-005 0.57
- 58 serafim-100m-portuguese-pt-sentence-encoder-ir 0.57
- 59 serafim-335m-portuguese-pt-sentence-encoder-ir 0.57
- 60 multilingual-e5-small 0.56
- 61 granite-embedding-107m-multilingual 0.56
- 62 granite-embedding-97m-multilingual-r2 0.56
- 63 serafim-900m-portuguese-pt-sentence-encoder-ir 0.56
- 64 serafim-335m-portuguese-pt-sentence-encoder 0.55
- 65 serafim-900m-portuguese-pt-sentence-encoder 0.54
- 66 F2LLM-v2-160M 0.54
- 67 paraphrase-multilingual-mpnet-base-v2 0.53
- 68 serafim-100m-portuguese-pt-sentence-encoder 0.53
- 69 Qwen3-Embedding-0.6B-jua-V2 0.52
- 70 F2LLM-v2-80M 0.51
- 71 LaBSE 0.50
- 72 medlink-bi-encoder 0.50
- 73 e5-small-v2 0.47
- 74 mxbai-embed-large-v1 0.47
- 75 bge-small-en-v1.5 0.47
- 76 bert-large-portuguese-cased 0.47
- 77 gte-small 0.46
- 78 bert-base-portuguese-cased 0.46
- 79 bert-large-portuguese-cased-legal-mlm-sts-v1.0 0.46
- 80 e5-mistral-7b-instruct 0.45
- 81 llama-embed-nemotron-8b 0.45
- 82 Legal-BERTimbau-sts-large 0.44
- 83 bert-large-portuguese-cased-legal-mlm-mkd-nli-sts-v1 0.44
- 84 Ivysaur 0.44
- 85 albertina-900m-portuguese-ptbr-encoder 0.43
- 86 all-MiniLM-L12-v2 0.43
- 87 GIST-all-MiniLM-L6-v2 0.42
- 88 all-mpnet-base-v2 0.42
- 89 legal-bert-pt-br 0.41
- 90 all-MiniLM-L6-v2 0.41
- 91 Legal-BERTimbau-sts-large-ma-v3 0.41
- 92 bert-large-portuguese-cased-legal-tsdae-gpl-nli-sts-MetaKD-v0 0.36
- 93 paraphrase-multilingual-MiniLM-L12-v2 0.25
Source: Official MTEB-BR leaderboard
- 1
Qwen3-Embedding-8B 0.67
- 2
Qwen3-Embedding-4B 0.66
- 3
Gemini Embedding 2 0.63
- 4
Qwen3-Embedding-0.6B 0.62
Source: Official MTEB-BR leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | gemini-embedding-001 closed weights | 0.68 | Official leaderboard | ||
| 2 | | 0.67 | Jun 2025 | Official leaderboard | |
| 3 | KaLM-Embedding-Gemma3-12B-2511 open weights, 11.8B | 0.67 | Official leaderboard | ||
| 4 | voyage-context-4 closed weights | 0.67 | Official leaderboard | ||
| 5 | Octen-Embedding-8B open weights, 7.6B | 0.67 | Official leaderboard | ||
| 6 | | 0.66 | Jun 2025 | Official leaderboard | |
| 7 | voyage-context-3 closed weights | 0.66 | Official leaderboard | ||
| 8 | voyage-3-large closed weights | 0.66 | Official leaderboard | ||
| 9 | voyage-4-large closed weights | 0.65 | Official leaderboard | ||
| 10 | SFR-Embedding-Mistral open weights, 7.1B | 0.65 | Official leaderboard | ||
| 11 | BidirLM-1.7B-Embedding open weights, 1.7B | 0.65 | Official leaderboard | ||
| 12 | BOOM_4B_v1 open weights, 4B | 0.65 | Official leaderboard | ||
| 13 | embeddinggemma-300m open weights, 0.3B | 0.65 | Official leaderboard | ||
| 14 | codestral-embed closed weights | 0.65 | Official leaderboard | ||
| 15 | Linq-Embed-Mistral open weights, 7.1B | 0.65 | Official leaderboard | ||
| 16 | text-embedding-3-large closed weights | 0.64 | Official leaderboard | ||
| 17 | jina-embeddings-v5-text-small open weights, 0.6B | 0.64 | Official leaderboard | ||
| 18 | voyage-4 closed weights | 0.64 | Official leaderboard | ||
| 19 | multilingual-e5-large-instruct open weights, 0.6B | 0.64 | Official leaderboard | ||
| 20 | SFR-Embedding-2_R open weights, 7.1B | 0.64 | Official leaderboard | ||
| 21 | gte-Qwen2-7B-instruct open weights, 7.1B | 0.64 | Official leaderboard | ||
| 22 | harrier-oss-v1-27b open weights, 27B | 0.64 | Official leaderboard | ||
| 23 | BidirLM-1B-Embedding open weights, 1B | 0.64 | Official leaderboard | ||
| 24 | F2LLM-v2-8B open weights, 7.6B | 0.64 | Official leaderboard | ||
| 25 | voyage-3.5 closed weights | 0.64 | Official leaderboard | ||
| 26 | harrier-oss-v1-0.6b open weights, 0.6B | 0.63 | Official leaderboard | ||
| 27 | F2LLM-v2-14B open weights, 14B | 0.63 | Official leaderboard | ||
| 28 | | 0.63 | Mar 2026 | Official leaderboard | |
| 29 | F2LLM-v2-4B open weights, 4B | 0.63 | Official leaderboard | ||
| 30 | PwC-Embedding_expr open weights, 0.6B | 0.63 | Official leaderboard | ||
| 31 | voyage-finance-2 closed weights | 0.63 | Official leaderboard | ||
| 32 | Octen-Embedding-0.6B open weights, 0.6B | 0.63 | Official leaderboard | ||
| 33 | PIXIE-Rune-v1.0 open weights, 0.6B | 0.62 | Official leaderboard | ||
| 34 | | 0.62 | Jun 2025 | Official leaderboard | |
| 35 | gte-Qwen2-1.5B-instruct open weights, 1.5B | 0.62 | Official leaderboard | ||
| 36 | voyage-law-2 closed weights | 0.62 | Official leaderboard | ||
| 37 | snowflake-arctic-embed-l-v2.0 open weights, 0.6B | 0.62 | Official leaderboard | ||
| 38 | voyage-3.5-lite closed weights | 0.62 | Official leaderboard | ||
| 39 | text-embedding-3-small closed weights | 0.62 | Official leaderboard | ||
| 40 | bge-m3 open weights, 0.6B | 0.62 | Official leaderboard | ||
| 41 | voyage-4-lite closed weights | 0.62 | Official leaderboard | ||
| 42 | F2LLM-v2-1.7B open weights, 1.7B | 0.61 | Official leaderboard | ||
| 43 | BidirLM-0.6B-Embedding open weights, 0.6B | 0.61 | Official leaderboard | ||
| 44 | voyage-3 closed weights | 0.61 | Official leaderboard | ||
| 45 | jua-4B-mixed open weights, 4B | 0.61 | Official leaderboard | ||
| 46 | F2LLM-0.6B open weights, 0.6B | 0.61 | Official leaderboard | ||
| 47 | harrier-oss-v1-270m open weights, 0.3B | 0.60 | Official leaderboard | ||
| 48 | F2LLM-v2-0.6B open weights, 0.6B | 0.60 | Official leaderboard | ||
| 49 | mistral-embed closed weights | 0.60 | Official leaderboard | ||
| 50 | titan-embed-text-v2 closed weights | 0.59 | Official leaderboard | ||
| 51 | multilingual-e5-large open weights, 0.6B | 0.59 | Official leaderboard | ||
| 52 | granite-embedding-311m-multilingual-r2 open weights, 0.3B | 0.59 | Official leaderboard | ||
| 53 | jua-4B-legal-only open weights, 4B | 0.59 | Official leaderboard | ||
| 54 | voyage-3-lite closed weights | 0.59 | Official leaderboard | ||
| 55 | F2LLM-v2-330M open weights, 0.3B | 0.58 | Official leaderboard | ||
| 56 | multilingual-e5-base open weights, 0.3B | 0.58 | Official leaderboard | ||
| 57 | text-embedding-005 closed weights | 0.57 | Official leaderboard | ||
| 58 | serafim-100m-portuguese-pt-sentence-encoder-ir open weights, 0.1B | 0.57 | Official leaderboard | ||
| 59 | serafim-335m-portuguese-pt-sentence-encoder-ir open weights, 0.3B | 0.57 | Official leaderboard | ||
| 60 | multilingual-e5-small open weights, 0.1B | 0.56 | Official leaderboard | ||
| 61 | granite-embedding-107m-multilingual open weights, 0.1B | 0.56 | Official leaderboard | ||
| 62 | granite-embedding-97m-multilingual-r2 open weights, 0.1B | 0.56 | Official leaderboard | ||
| 63 | serafim-900m-portuguese-pt-sentence-encoder-ir open weights, 0.9B | 0.56 | Official leaderboard | ||
| 64 | serafim-335m-portuguese-pt-sentence-encoder open weights, 0.3B | 0.55 | Official leaderboard | ||
| 65 | serafim-900m-portuguese-pt-sentence-encoder open weights, 0.9B | 0.54 | Official leaderboard | ||
| 66 | F2LLM-v2-160M open weights, 0.2B | 0.54 | Official leaderboard | ||
| 67 | paraphrase-multilingual-mpnet-base-v2 open weights, 0.3B | 0.53 | Official leaderboard | ||
| 68 | serafim-100m-portuguese-pt-sentence-encoder open weights, 0.1B | 0.53 | Official leaderboard | ||
| 69 | Qwen3-Embedding-0.6B-jua-V2 open weights, 0.6B | 0.52 | Official leaderboard | ||
| 70 | F2LLM-v2-80M open weights, 0.1B | 0.51 | Official leaderboard | ||
| 71 | LaBSE open weights, 0.5B | 0.50 | Official leaderboard | ||
| 72 | medlink-bi-encoder open weights, 0.1B | 0.50 | Official leaderboard | ||
| 73 | e5-small-v2 open weights, 0B | 0.47 | Official leaderboard | ||
| 74 | mxbai-embed-large-v1 open weights, 0.3B | 0.47 | Official leaderboard | ||
| 75 | bge-small-en-v1.5 open weights, 0B | 0.47 | Official leaderboard | ||
| 76 | bert-large-portuguese-cased open weights, 0.3B | 0.47 | Official leaderboard | ||
| 77 | gte-small open weights, 0B | 0.46 | Official leaderboard | ||
| 78 | bert-base-portuguese-cased open weights, 0.1B | 0.46 | Official leaderboard | ||
| 79 | bert-large-portuguese-cased-legal-mlm-sts-v1.0 open weights, 0.3B | 0.46 | Official leaderboard | ||
| 80 | e5-mistral-7b-instruct open weights, 7.1B | 0.45 | Official leaderboard | ||
| 81 | llama-embed-nemotron-8b open weights, 7.5B | 0.45 | Official leaderboard | ||
| 82 | Legal-BERTimbau-sts-large open weights, 0.3B | 0.44 | Official leaderboard | ||
| 83 | bert-large-portuguese-cased-legal-mlm-mkd-nli-sts-v1 open weights, 0.3B | 0.44 | Official leaderboard | ||
| 84 | Ivysaur open weights, 0B | 0.44 | Official leaderboard | ||
| 85 | albertina-900m-portuguese-ptbr-encoder open weights, 0.9B | 0.43 | Official leaderboard | ||
| 86 | all-MiniLM-L12-v2 open weights, 0B | 0.43 | Official leaderboard | ||
| 87 | GIST-all-MiniLM-L6-v2 open weights, 0B | 0.42 | Official leaderboard | ||
| 88 | all-mpnet-base-v2 open weights, 0.1B | 0.42 | Official leaderboard | ||
| 89 | legal-bert-pt-br open weights, 0.1B | 0.41 | Official leaderboard | ||
| 90 | all-MiniLM-L6-v2 open weights, 0B | 0.41 | Official leaderboard | ||
| 91 | Legal-BERTimbau-sts-large-ma-v3 open weights, 0.3B | 0.41 | Official leaderboard | ||
| 92 | bert-large-portuguese-cased-legal-tsdae-gpl-nli-sts-MetaKD-v0 open weights, 0.3B | 0.36 | Official leaderboard | ||
| 93 | paraphrase-multilingual-MiniLM-L12-v2 open weights, 0.1B | 0.25 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 0.67 | Jun 2025 | Official leaderboard | |
| 2 | | 0.66 | Jun 2025 | Official leaderboard | |
| 3 | | 0.63 | Mar 2026 | Official leaderboard | |
| 4 | | 0.62 | Jun 2025 | Official leaderboard |
Rankings by part
We read 22 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as HateBR, FactckBrClassification and ToxSynPT. Gemini Embedding 2 leads 9 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
HateBR
4 models we track have a score here.
- 1
Gemini Embedding 2 ↑2 0.85
- 2
Qwen3-Embedding-8B ↓1 0.84
- 3
Qwen3-Embedding-4B ↓1 0.82
- 4
Qwen3-Embedding-0.6B 0.78
FactckBrClassification
4 models we track have a score here.
- 1
Gemini Embedding 2 ↑2 0.57
- 2
Qwen3-Embedding-8B ↓1 0.54
- 3
Qwen3-Embedding-0.6B ↑1 0.51
- 4
Qwen3-Embedding-4B ↓2 0.51
ToxSynPT
4 models we track have a score here.
- 1
Gemini Embedding 2 ↑2 0.90
- 2
Qwen3-Embedding-8B ↓1 0.87
- 3
Qwen3-Embedding-0.6B ↑1 0.84
- 4
Qwen3-Embedding-4B ↓2 0.83
PortuLexRRIP
4 models we track have a score here.
- 1
Gemini Embedding 2 ↑2 0.49
- 2
Qwen3-Embedding-8B ↓1 0.35
- 3
Qwen3-Embedding-0.6B ↑1 0.33
- 4
Qwen3-Embedding-4B ↓2 0.33
BrighterEmotionMultilabelClassification
4 models we track have a score here.
- 1
Qwen3-Embedding-4B ↑1 0.29
- 2
Qwen3-Embedding-8B ↓1 0.28
- 3
Qwen3-Embedding-0.6B ↑1 0.27
- 4
Gemini Embedding 2 ↓1 0.25
AssinRTE
4 models we track have a score here.
- 1
Qwen3-Embedding-8B 0.87
- 2
Qwen3-Embedding-4B 0.87
- 3
Qwen3-Embedding-0.6B ↑1 0.83
- 4
Gemini Embedding 2 ↓1 0.77
InferBR
4 models we track have a score here.
- 1
Qwen3-Embedding-8B 0.91
- 2
Qwen3-Embedding-4B 0.89
- 3
Qwen3-Embedding-0.6B ↑1 0.82
- 4
Gemini Embedding 2 ↓1 0.69
AssinSTS
4 models we track have a score here.
- 1
Qwen3-Embedding-4B ↑1 0.79
- 2
Qwen3-Embedding-8B ↓1 0.79
- 3
Gemini Embedding 2 0.75
- 4
Qwen3-Embedding-0.6B 0.74
Assin2STS
4 models we track have a score here.
- 1
Qwen3-Embedding-8B 0.83
- 2
Qwen3-Embedding-4B 0.82
- 3
Qwen3-Embedding-0.6B ↑1 0.78
- 4
Gemini Embedding 2 ↓1 0.73
MedPTClustering
4 models we track have a score here.
- 1
Qwen3-Embedding-4B ↑1 0.89
- 2
Gemini Embedding 2 ↑1 0.74
- 3
Qwen3-Embedding-8B ↓2 0.72
- 4
Qwen3-Embedding-0.6B 0.70
WikipediaPTCategoriesClusteringP2P
4 models we track have a score here.
- 1
Qwen3-Embedding-8B 0.80
- 2
Qwen3-Embedding-0.6B ↑2 0.77
- 3
Qwen3-Embedding-4B ↓1 0.73
- 4
Gemini Embedding 2 ↓1 0.61
JurisTCUClusteringP2P
4 models we track have a score here.
- 1
Qwen3-Embedding-4B ↑1 0.41
- 2
Qwen3-Embedding-8B ↓1 0.37
- 3
Qwen3-Embedding-0.6B ↑1 0.37
- 4
Gemini Embedding 2 ↓1 0.29
SciELOClusteringP2P
4 models we track have a score here.
- 1
Qwen3-Embedding-8B 0.78
- 2
Qwen3-Embedding-4B 0.74
- 3
Gemini Embedding 2 0.73
- 4
Qwen3-Embedding-0.6B 0.71
StackoverflowPtClustering
4 models we track have a score here.
- 1
Qwen3-Embedding-4B ↑1 0.63
- 2
Qwen3-Embedding-8B ↓1 0.58
- 3
Qwen3-Embedding-0.6B ↑1 0.55
- 4
Gemini Embedding 2 ↓1 0.49
MedPTRetrieval
4 models we track have a score here.
- 1
Gemini Embedding 2 ↑2 0.88
- 2
Qwen3-Embedding-8B ↓1 0.82
- 3
Qwen3-Embedding-4B ↓1 0.77
- 4
Qwen3-Embedding-0.6B 0.69
FaQuADIR
4 models we track have a score here.
- 1
Gemini Embedding 2 ↑2 0.86
- 2
Qwen3-Embedding-8B ↓1 0.84
- 3
Qwen3-Embedding-4B ↓1 0.81
- 4
Qwen3-Embedding-0.6B 0.79
Quati
4 models we track have a score here.
- 1
Gemini Embedding 2 ↑2 0.65
- 2
Qwen3-Embedding-8B ↓1 0.64
- 3
Qwen3-Embedding-4B ↓1 0.61
- 4
Qwen3-Embedding-0.6B 0.56
FaqBacenRetrieval
4 models we track have a score here.
- 1
Gemini Embedding 2 ↑2 0.78
- 2
Qwen3-Embedding-8B ↓1 0.71
- 3
Qwen3-Embedding-4B ↓1 0.65
- 4
Qwen3-Embedding-0.6B 0.63
JurisTCU
4 models we track have a score here.
- 1
Qwen3-Embedding-8B 0.62
- 2
Qwen3-Embedding-4B 0.62
- 3
Qwen3-Embedding-0.6B ↑1 0.57
- 4
Gemini Embedding 2 ↓1 0.49
BRTaxQAR
4 models we track have a score here.
- 1
Qwen3-Embedding-8B 0.42
- 2
Qwen3-Embedding-4B 0.42
- 3
Qwen3-Embedding-0.6B ↑1 0.37
- 4
Gemini Embedding 2 ↓1 0.29
QuatiReranking
4 models we track have a score here.
- 1
Gemini Embedding 2 ↑2 0.64
- 2
Qwen3-Embedding-8B ↓1 0.63
- 3
Qwen3-Embedding-4B ↓1 0.60
- 4
Qwen3-Embedding-0.6B 0.57
JurisTCUReranking
4 models we track have a score here.
- 1
Qwen3-Embedding-8B 0.56
- 2
Qwen3-Embedding-4B 0.55
- 3
Qwen3-Embedding-0.6B ↑1 0.53
- 4
Gemini Embedding 2 ↓1 0.48
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official MTEB-BR leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official MTEB-BR leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Multilingual
- MMMLU In use MMLU professionally translated into 14 languages. Multiple-choice accuracy. Claude Mythos Preview 92.7%
- WMT24++ In use English-to-55-language translation of WMT24 test data. Scored with automatic metrics such as MetricX or xCOMET.
K-EXAONE 90.5%
- MMLU-ProX In use MMLU-Pro translated into up to 29 languages with expert review. Multiple-choice accuracy per language.
Qwen3.7-Max 87%
- CLIcK In use Models answer 1,995 multiple-choice questions on Korean language and culture. Questions come from official Korean exams and textbooks in eleven categories.
Solar Open 2 90.7%
- Global MMLU Lite 3.0 In use Models answer 400 human-translated MMLU questions per language in 23 languages. Version 3.0 adds Czech, Hungarian, Oriya, Slovak and Tajik and updates Italian. Accuracy. Claude Opus 5 95.7%
- MMMB In use Models answer multiple-choice questions about images in six languages across 15 categories. The set has about 12,000 questions. LFM2.5-VL-3B 83%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 8 months ago, in March 2026, for Gemini Embedding 2. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (93 entries).
- Pass Models scored 4 models we track, plus 89 other entries on its leaderboard.
- Warning Newest score The newest came 8 months ago, in March 2026, for Gemini Embedding 2.
- Note Frontier models None of its models are on the capabilities index.
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on huggingface.co: 4 rows for models we track and 89 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.