NVIDIA: Nemotron 3.5 Embed 1B (free) vs Sentence Transformers: all-mpnet-base-v2
A side-by-side comparison of NVIDIA: Nemotron 3.5 Embed 1B (free) (Free) and Sentence Transformers: all-mpnet-base-v2 ($0.02 / 1M tokens) using current USD prices and capabilities from the api14 catalog. Both run through the same OpenAI-compatible API.
Specifications
| NVIDIA: Nemotron 3.5 Embed 1B (free) | Sentence Transformers: all-mpnet-base-v2 | |
|---|---|---|
| Provider | NVIDIA | Sentence Transformers |
| Input | - | $0.02 / 1M tokens |
| Context window | 32,768 | 512 |
| Max output | 29,491 | 460 |
| Input | text, image | text |
| Output | embeddings | embeddings |
| Tool calling | ||
| Structured output | ||
| Reasoning control | ||
| Added | Oct 21, 2026 | Nov 17, 2025 |
Key differences
- NVIDIA: Nemotron 3.5 Embed 1B (free) reads 64.0× more context (32,768 tokens).
In depth
NVIDIA: Nemotron 3.5 Embed 1B (free) is an embedding model from NVIDIA, listed under the model id nvidia/nemotron-3.5-embed-1b:free. Sentence Transformers: all-mpnet-base-v2 is an embedding model from Sentence Transformers, listed under the model id sentence-transformers/all-mpnet-base-v2. Both sit in the embeddings category, so neither one generates chat completions or free text; each takes input and returns a vector representation you can use for search, clustering, deduplication or classification. The difference in who makes them matters mainly for how you expect them to behave on your data, since NVIDIA ships a newer model while all-mpnet-base-v2 is a long established Sentence Transformers checkpoint.
The clearest technical gap is context length. Nemotron 3.5 Embed 1B accepts up to 32768 tokens, while all-mpnet-base-v2 accepts 512 tokens. That is a large difference in how much text you can embed in a single call before you need to chunk. Nemotron 3.5 Embed 1B also accepts text and image input and returns embeddings, so it is multimodal on the input side. all-mpnet-base-v2 accepts text only and returns embeddings. Neither model exposes supported parameters in the data given here, so you should not expect tunable knobs such as dimensions or pooling options to be documented for either one. The qualitative comparison marks Nemotron 3.5 Embed 1B as the one with the bigger context, and the price fields are unknown, so treat them as roughly comparable until you check the live table.
For practical selection, reach for Nemotron 3.5 Embed 1B when your documents are long, when you want to embed an entire page, transcript or report without splitting it first, or when your input includes images alongside text. The 32768 token window lets you avoid aggressive chunking, which often preserves meaning better for retrieval. Reach for all-mpnet-base-v2 when your inputs are short, such as sentences, product titles, short queries or individual paragraphs, and when you want a plain text-only embedding pipeline with no image handling. If your corpus is already chunked to a few hundred tokens, the smaller context window is not a limitation, and the simpler text-only interface may be easier to reason about.
Calling either model through api14 works the same way because the API is OpenAI-compatible. You send your input to the embeddings endpoint with your api14 key, and the only thing you change between the two is the model id string. Use nvidia/nemotron-3.5-embed-1b:free for the NVIDIA model and sentence-transformers/all-mpnet-base-v2 for the Sentence Transformers model. The request shape, the response shape and the way you read the returned vectors stay the same, so you can swap the id in a config value or environment variable and compare results on your own data without rewriting client code.
There is a case where all-mpnet-base-v2 remains the better pick even though it has the smaller window. If all of your inputs are short text and you have no images, the extra context and multimodal input of Nemotron 3.5 Embed 1B buy you nothing, and a text-only model keeps your pipeline and your evaluation simpler. It is also the natural choice if you already have tooling, thresholds or index settings tuned around its 512 token behavior and you do not want to re-chunk your corpus.
Switch with one string
Both models use the same endpoint and request shape. Change the model value from "nvidia/nemotron-3.5-embed-1b:free" to "sentence-transformers/all-mpnet-base-v2" and prompts, streaming and tools keep working.
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api14.com/api/v1",
apiKey: process.env.API14_API_KEY,
});
const result = await client.embeddings.create({
"model": "sentence-transformers/all-mpnet-base-v2",
"input": [
"The quick brown fox",
"Lorem ipsum"
]
});
console.log(result.data.map((d) => d.embedding.length));
More comparisons
- Amazon: Titan Multimodal Embeddings G1 vs Amazon: Titan Embeddings G1 - Text
- Amazon: Titan Multimodal Embeddings G1 vs Amazon: Titan Text Embeddings V2
- Amazon: Titan Multimodal Embeddings G1 vs Cohere: Embed 5 Fast
- Amazon: Titan Multimodal Embeddings G1 vs Cohere: Embed 5 Pro
- Amazon: Titan Embeddings G1 - Text vs Amazon: Titan Text Embeddings V2
- Amazon: Titan Embeddings G1 - Text vs Cohere: Embed 5 Fast
- Amazon: Titan Embeddings G1 - Text vs Cohere: Embed 5 Pro
- Amazon: Titan Text Embeddings V2 vs Cohere: Embed 5 Fast