SAVRN Model Hub · Comparisons
all-MiniLM-L6-v2 vs multi-qa-MiniLM-L6-cos-v1
All-MiniLM-L6-v2 has 23M parameters and multi-qa-MiniLM-L6-cos-v1 has 23M parameters; at 16-bit, all-MiniLM-L6-v2 needs about 0.1 GB (1x MI300X from $1.85 an hour) and multi-qa-MiniLM-L6-cos-v1 about 0.1 GB (1x MI300X from $1.85 an hour).
| Field | all-MiniLM-L6-v2 sentence-transformers/all-MiniLM-L6-v2 | multi-qa-MiniLM-L6-cos-v1 sentence-transformers/multi-qa-MiniLM-L6-cos-v1 |
|---|---|---|
| Publisher | Sentence Transformers | Sentence Transformers |
| Task | Sentence similarity | Sentence similarity |
| Modality | Text | Text |
| Parameters, as reported | 23M parameters | 23M parameters |
| Architecture | BertModel | BertModel |
| Library | sentence-transformers | sentence-transformers |
| Context length | 512 tokens | 512 tokens |
| Repository size | 976.9 MB | 886.0 MB |
| Artifact formats | safetensors, onnx, openvino, pytorch, rust, tf | safetensors, onnx, openvino, pytorch, tf |
| License | apache-2.0 | Not stated |
| Access | Open weights, no gate | Open weights, no gate |
| Memory at 16-bit (weights and margin) | 0.1 GB | 0.1 GB |
| Cheapest GPUs at 16-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Memory at 4-bit (weights and margin) | 0 GB | 0 GB |
| Cheapest GPUs at 4-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Revision viewed | 1110a243fdf4 | b20736733232 |
| Downloads reported by the hub | 255.1M | 797.5k |
| 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.
Other Reported Results
These results are listed for each model on its own, because the conditions needed to compare them are not stated or do not match. Two results that leave a condition blank are not assumed to share it.
all-MiniLM-L6-v2
| Benchmark | Conditions | Result | Reported by | Revision | Date |
|---|---|---|---|---|---|
| mteb/arguana | Task ArguAnaMetric ArguAnaSetup Obtained using MTEB v1.12.75Comparison conditions not established | 50.167 | Obtained using MTEB v1.12.75 Reported by a third party |
Evaluated revision not stated | 2026-03-05 |
| mteb/arguana | Task ArguAna_default_testMetric ArguAna_default_testSetup Obtained using MTEB v1.12.75Comparison conditions not established | 50.167 | Obtained using MTEB v1.12.75 Reported by a third party |
Evaluated revision not stated | 2026-03-05 |
SAVRN's Notes on all-MiniLM-L6-v2
Sentence embedding sits at the front of almost every retrieval pipeline we run, and the memory line tells you where this model belongs: 0.1 GB at 16-bit, rounding to zero at 8-bit and 4-bit. The cheapest on-demand setup we price is one MI300X with 192 GB at $1.85 an hour, so nobody rents a card for this alone. It rides in the spare memory of whatever GPU already serves your main model, 23M parameters producing a 384-wide vector.
Apache 2.0 lets you fine-tune it and ship the result commercially, provided the license, copyright and NOTICE files travel with it and significant changes are marked. Check the 512-token context against your document lengths, since longer text gets split first, and read the lineage: nreimers/MiniLM-L6-H384-uncased as the source model and seven training sets, including ms_marco, s2orc and code_search_net. No host prices it by the token in our Index.
SAVRN's Notes on multi-qa-MiniLM-L6-cos-v1
Questions on one side, answers on the other: that pairing is what Sentence Transformers trained this 23M-parameter embedder on, 215M question-and-answer pairs from sources including ms_marco, gooaq, natural_questions, trivia_qa and eli5. Six layers, a 384-dimensional output, built for semantic search. Memory is 0.1 GB at 16-bit and rounds to 0.0 GB at 8-bit and 4-bit, so the cheapest host we list, one MI300X with 192 GB at $1.85 an hour on-demand, is a card this model shares with whatever generator answers the query.
The license field is blank. Access is open, so the files come from the publisher without a gate, but blank is not permission, and a deployment that earns money needs the terms in writing before it ships. Check the 512-token context against the length of the passages you index. Released March 2, 2022, last updated November 5, 2024, and no Index host prices it per token.
Questions
Which is larger, all-MiniLM-L6-v2 or multi-qa-MiniLM-L6-cos-v1?
all-MiniLM-L6-v2 (23M parameters) is larger than multi-qa-MiniLM-L6-cos-v1 (23M parameters), by the parameter counts their publishers report.
Which is cheaper to run, all-MiniLM-L6-v2 or multi-qa-MiniLM-L6-cos-v1?
At 4-bit, all-MiniLM-L6-v2 fits on 1x MI300X from $1.85 an hour and multi-qa-MiniLM-L6-cos-v1 on 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use all-MiniLM-L6-v2 commercially?
Yes. all-MiniLM-L6-v2 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.