SAVRN
Search Contact SAVRN

SAVRN Model Hub · Comparisons

llama-nemotron-rerank-1b-v2 vs Qwen3-Reranker-4B

Llama-nemotron-rerank-1b-v2 has 1.2B parameters and Qwen3-Reranker-4B has 4B parameters; llama-nemotron-rerank-1b-v2 is released under other and Qwen3-Reranker-4B under Apache License 2.0; at 16-bit, llama-nemotron-rerank-1b-v2 needs about 3 GB (1x MI300X from $1.85 an hour) and Qwen3-Reranker-4B about 9.7 GB (1x MI300X from $1.85 an hour).

Published metadata for 2 models, each read from its own repository.
Field llama-nemotron-rerank-1b-v2
nvidia/llama-nemotron-rerank-1b-v2
Qwen3-Reranker-4B
Qwen/Qwen3-Reranker-4B
Publisher NVIDIA Qwen
Task Text ranking Text ranking
Modality Other Other
Parameters, as reported 1.2B parameters 4B parameters
Architecture LlamaBidirectionalForSequenceClassification Qwen3ForCausalLM
Library transformers transformers
Context length 131,072 tokens 40,960 tokens
Repository size 5.0 GB 8.1 GB
Artifact formats safetensors, pytorch safetensors
License other apache-2.0
Access Open weights, no gate Open weights, no gate
Memory at 16-bit (weights and margin) 3 GB 9.7 GB
Cheapest GPUs at 16-bit, per hour 1x MI300X, $1.85 1x MI300X, $1.85
Memory at 4-bit (weights and margin) 0.7 GB 2.4 GB
Cheapest GPUs at 4-bit, per hour 1x MI300X, $1.85 1x MI300X, $1.85
Revision viewed 828765652b05 22e683669bc0
Downloads reported by the hub 981.9k 2.5M
Last observed 2026-09-18 2026-09-18

An evaluation row appears only where at least two of these models report the same benchmark with the same stated configuration, metric, unit and setup. Different evaluators stay named in each cell. Values are shown as reported: no unit conversion, no ranking.

SAVRN's Notes on llama-nemotron-rerank-1b-v2

A reranker sits between the retriever and the generator, and this one returns a logit score for how relevant a document is to a query. NVIDIA fine-tuned it for multilingual and cross-lingual question-answering retrieval and evaluated it across 26 languages, with document support up to 8,192 tokens even though the configuration allows 131,072 positions. At 1.2B parameters the 16-bit weights take 2.5 GB and the run needs 3.0 GB; 4-bit brings that to 0.7 GB. The cheapest priced setup, one MI300X with 192 GB at $1.85 an hour on-demand, is far more card than this needs, so co-locate it with the generator.

The license is listed only as other, with no summary, so read the publisher's terms yourself before anything commercial goes live. Chunk to the 8,192-token document limit, not the config figure, and note the last update was 2026-08-26, so pin the revision you validated.

SAVRN's Notes on Qwen3-Reranker-4B

Second stage only. A reranker rescores the candidates an embedding search already pulled back, and every query-document pair is a full pass through 36 layers. That makes it compute per query rather than a memory problem: 8.0 GB of weights at 16-bit, 9.7 GB to run, 2.4 GB at 4-bit. On the cheapest setup we list, one 192 GB MI300X at $1.85 an hour, memory never binds; pairs scored per second does, so size for traffic. The 40,960-token context caps query plus document.

Nothing in Apache 2.0 stops you shipping it in a product; keep the notices, state significant changes, plus the patent grant. Check lineage and dates. It is derived from Qwen3-4B-Base, on our hub, and shares one paper with the series' embedding models, arXiv:2506.05176. Released June 3, 2025, files last updated April 16, 2026, ten months on; diff what changed before rebuilding a serving image.

Questions

Which is larger, llama-nemotron-rerank-1b-v2 or Qwen3-Reranker-4B?

Qwen3-Reranker-4B (4B parameters) is larger than llama-nemotron-rerank-1b-v2 (1.2B parameters), by the parameter counts their publishers report.

Which is cheaper to run, llama-nemotron-rerank-1b-v2 or Qwen3-Reranker-4B?

At 4-bit, llama-nemotron-rerank-1b-v2 fits on 1x MI300X from $1.85 an hour and Qwen3-Reranker-4B on 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use Qwen3-Reranker-4B commercially?

Yes. Qwen3-Reranker-4B is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

Related Comparisons