Choose embedding models for your retrieval task.
Choose an embedding model against the material you need to retrieve. This view brings listed input context, an available input-token quote and a named embedding benchmark together, while keeping missing evidence visible.
63 embedding models are recorded. Up to 12 appear, with measured models first and then larger listed context. Scores shown all use FreshStack Retrieval NDCG@10 (MTEB 1.38.43; test/default; dataset 7a20df1a).
Catalog snapshot · 12 rows shownMatch the retrieval task
An embedding model maps input into vectors used for search and retrieval. The right check is whether your queries retrieve the passages you need. Test with the language, document length and kinds of near misses your application will encounter.
A benchmark aggregate does not establish performance on every retrieval task. When a retained MTEB metric is available, all scores shown here use that metric and version. Open its source for the dataset and setup before using it as a shortlist signal. Unscored rows have no inferred score.
Check the integration before re-indexing
Confirm the embedding dimensions, supported input types, task prompts and normalization in the provider documentation. Those fields are not structured catalog facts here, so the table does not guess them from descriptions. A listed context window is the reported input capacity, not a recommended chunk size.
Keep indexing and query embeddings compatible. Changing model or vector dimensions can require a new index. For cost, use input-token billing from the actual serving route and include both your initial corpus and later updates. No current eligible quote means the cost remains unknown in this view.
How this view is assembled
Selection: at most 12 catalog embedding models, with results for one retained MTEB metric first, then listed context and name. A quoted input rate comes from one current eligible million-token route, chosen by provider order rather than a price winner. Scores, quotes and catalog specifications retain their separate sources and check dates.
Embedding evidence for a retrieval shortlist
Benchmark coverage first, then listed context; this order is not a quality ranking
| Model | Input context | Input API rate | Benchmark evidence | Sources and dates |
|---|---|---|---|---|
| Qwen3-Embedding-8B | 41K tokens | $0.01 / 1M input tokensOpenRouterConfiguration not reportedUSD | 0.43017FreshStack Retrieval NDCG@10MTEB 1.38.43; test/default; dataset 7a20df1aRevision 4e423935; test/defaultContributor reported | OpenRouter ↗Catalog updated DeepInfra ↗Catalog updated OpenRouter price source ↗Checked Benchmark result ↗Checked |
| NVIDIA: Llama Nemotron Embed VL 1B V2 (free) | 131.1K tokens | $0.00 / 1M input tokensOpenRouterConfiguration not reportedUSD | Not reported for the selected metric | OpenRouter ↗Catalog updated OpenRouter price source ↗Checked |
| Nemotron-3-Embed-1B-BF16 | 32.8K tokens | No current eligible input quote | Not reported for the selected metric | DeepInfra ↗Catalog updated |
| Nemotron-3-Embed-1B-NVFP4 | 32.8K tokens | No current eligible input quote | Not reported for the selected metric | DeepInfra ↗Catalog updated |
| Nemotron-3-Embed-8B | 32.8K tokens | No current eligible input quote | Not reported for the selected metric | DeepInfra ↗Catalog updated |
| Nemotron-3.5-Embed-1B-BF16 | 32.8K tokens | No current eligible input quote | Not reported for the selected metric | DeepInfra ↗Catalog updated |
| Nemotron-3.5-Embed-8B-BF16 | 32.8K tokens | No current eligible input quote | Not reported for the selected metric | DeepInfra ↗Catalog updated |
| NVIDIA: Nemotron 3 Embed 1B (free) | 32.8K tokens | $0.00 / 1M input tokensOpenRouterConfiguration not reportedUSD | Not reported for the selected metric | OpenRouter ↗Catalog updated OpenRouter price source ↗Checked |
| NVIDIA: Nemotron 3.5 Embed 1B (free) | 32.8K tokens | $0.00 / 1M input tokensOpenRouterConfiguration not reportedUSD | Not reported for the selected metric | OpenRouter ↗Catalog updated OpenRouter price source ↗Checked |
| NVIDIA: Nemotron 3.5 Embed 8B (free) | 32.8K tokens | $0.00 / 1M input tokensOpenRouterConfiguration not reportedUSD | Not reported for the selected metric | OpenRouter ↗Catalog updated OpenRouter price source ↗Checked |
| Qwen3-Embedding-0.6B | 32.8K tokens | No current eligible input quote | Not reported for the selected metric | DeepInfra ↗Catalog updated |
| Qwen3-Embedding-4B | 32.8K tokens | $0.02 / 1M input tokensOpenRouterConfiguration not reportedUSD | Not reported for the selected metric | OpenRouter ↗Catalog updated DeepInfra ↗Catalog updated OpenRouter price source ↗Checked |
Where these rows come from
We join catalog specifications, recorded provider quotes and published evaluation records where exact model identities are available. Some observations come through aggregators or benchmark collections. Multiple listings do not establish independent confirmation of a claim.
Open the source beside a row to check the original evidence. A collection or check date is separate from a model release or evaluation date. Unknown values stay unknown. Recorded prices are in USD with their original billing units and settings.
See data coverage and sources →Compare for your own use case
Keep your task, model identity and provider configuration together. Explore the catalog yourself, or connect your AI to read the same source-backed records.