Mixedbread Quality Evals
How Mixedbread performs on public retrieval and agent benchmarks, measured against the strongest competing models.
Every result comes from a public benchmark on its standard split and evaluation script. Retrieval numbers are Wholembed V3 served through the Stores API; agent numbers come from Toast 1 or third-party agents running on Mixedbread Search, with cost and latency measured per query at list prices. Each benchmark lists its dataset, exact methodology and testing dates.
How we measure
Every result comes from a public benchmark on its standard split and evaluation script. Retrieval numbers are Wholembed V3 served through the Stores API; agent numbers come from Toast 1 or third-party agents running on Mixedbread Search, with cost and latency measured per query at list prices. Each benchmark lists its dataset, exact methodology and testing dates.
Wholembed V3, served through the Stores API, against other embedding models on public retrieval evals.
ViDoRe V3 Crosslingual (Markdown)
Open Introducing ViDoRe V3Real-world, domain-specific document retrieval in markdown, with cross-lingual queries (de, en, es, fr, it, pt).
Per-domain breakdown
| Model | CS (EN) Open CS (EN) dataset | Finance (EN) Open Finance (EN) dataset | Pharma (EN) Open Pharma (EN) dataset | HR (EN) Open HR (EN) dataset | Industrial (EN) Open Industrial (EN) dataset | Energy (FR) Open Energy (FR) dataset | Finance (FR) Open Finance (FR) dataset | Physics (FR) Open Physics (FR) dataset |
|---|---|---|---|---|---|---|---|---|
| Mixedbread | 76.8 | 64.0 | 64.9 | 61.9 | 48.5 | 65.6 | 52.2 | 47.1 |
| + mxbai-rerank-v3.1-listwise | 86.2 | 72.8 | 73.4 | 72.8 | 61.9 | 75.5 | 58.9 | 52.6 |
| Voyage 4 Large | 75.5 | 60.0 | 66.8 | 60.2 | 47.3 | 65.2 | 46.5 | 48.5 |
| Cohere Embed 4 | 69.0 | 57.2 | 62.0 | 51.1 | 41.7 | 59.7 | 43.0 | 44.5 |
| Qwen3 Embedding 8B | 72.3 | 51.4 | 60.0 | 50.3 | 40.8 | 59.0 | 36.0 | 45.9 |
| Gemini Embedding 2 | 66.9 | 59.5 | 59.4 | 54.7 | 40.6 | 48.9 | 39.7 | 43.1 |
| OpenAI text-embedding-3-large | 64.9 | 51.0 | 59.3 | 48.1 | 35.0 | 53.1 | 32.1 | 44.6 |
NDCG@10 · higher is betterTested September 2026Introducing ViDoRe V3
Dataset and methodology
8 specialized industry corpora (computer science, finance, pharma, HR, industrial, energy, physics) paired with queries in 6 languages: German, English, Spanish, French, Italian, Portuguese. Default view is the per-domain average across all 6 query languages; toggle a language to see scores for that query language alone.
We evaluate retrieval quality using publicly available eval datasets. Each model encodes the corpus and queries using its standard inference pipeline, then we compute the reported metric over the full test split. We report results as published by the eval or as measured on identical splits with default evaluation scripts.