VectorDBBench(VDBBench): A Benchmark Tool for VectorDB

What is VDBBench

VDBBench is not just an offering of benchmark results for mainstream vector databases and cloud services, it's your go-to tool for the ultimate performance and cost-effectiveness comparison. Designed with ease-of-use in mind, VDBBench is devised to help users, even non-professionals, reproduce results or test new systems, making the hunt for the optimal choice amongst a plethora of cloud services and open-source vector databases a breeze.

Understanding the importance of user experience, we provide an intuitive visual interface. This not only empowers users to initiate benchmarks at ease, but also to view comparative result reports, thereby reproducing benchmark results effortlessly. To add more relevance and practicality, we provide cost-effectiveness reports particularly for cloud services. This allows for a more realistic and applicable benchmarking process.

Closely mimicking real-world production environments, we've set up diverse testing scenarios including insertion, searching, and filtered searching. To provide you with credible and reliable data, we've included public datasets from actual production scenarios, such as SIFT, GIST, Cohere, and a dataset generated by OpenAI from an opensource raw dataset. It's fascinating to discover how a relatively unknown open-source database might excel in certain circumstances!

Prepare to delve into the world of VDBBench, and let it guide you in uncovering your perfect vector database match.

VDBBench is sponsored by Zilliz，the leading opensource vectorDB company behind Milvus. Choose smarter with VDBBench - start your free test on zilliz cloud today!

Leaderboard:https://zilliz.com/benchmark

🎈 Announcement 🎈

June 2026 update: Full Text Search has landed in VectorDBBench. We now benchmark BM25-style retrieval across supported backends, starting with MS MARCO and HotpotQA datasets, payload profiles, recall, QPS, and load metrics ready to compare. See the VectorDBBench Full Text Search Release Note for the full rollout details and caveats.

Quick Start

Prerequirement

python >= 3.11

Install

Install vectordb-bench with only PyMilvus

pip install vectordb-bench

Install the specific database client

pip install 'vectordb-bench[pinecone]'

All the database client supported

Optional database client	install command
pymilvus, zilliz_cloud (default)	pip install vectordb-bench
qdrant	pip install vectordb-bench[qdrant]
pinecone	pip install vectordb-bench[pinecone]
weaviate	pip install vectordb-bench[weaviate]
elastic, aliyun_elasticsearch	pip install vectordb-bench[elastic]
pgvector, pgvectorscale, pgdiskann, alloydb, vectorchord	pip install vectordb-bench[pgvector]
pgvecto.rs	pip install vectordb-bench[pgvecto_rs]
redis	pip install vectordb-bench[redis]
memorydb	pip install vectordb-bench[memorydb]
chromadb	pip install vectordb-bench[chromadb]
cockroachdb	pip install vectordb-bench[cockroachdb]
awsopensearch	pip install vectordb-bench[opensearch]
aliyun_opensearch	pip install vectordb-bench[aliyun_opensearch]
mongodb	pip install vectordb-bench[mongodb]
tidb	pip install vectordb-bench[tidb]
vespa	pip install vectordb-bench[vespa]
oceanbase	pip install vectordb-bench[oceanbase]
hologres	pip install vectordb-bench[hologres]
tencent_es	pip install vectordb-bench[tencent_es]
alisql	pip install vectordb-bench[alisql]
polardb	pip install vectordb-bench[polardb]
doris	pip install vectordb-bench[doris]
zvec	pip install vectordb-bench[zvec]
endee	pip install vectordb-bench[endee]
lindorm	pip install vectordb-bench[lindorm]
volc_mysql	pip install vectordb-bench[volc_mysql]
adbpg	pip install vectordb-bench[adbpg]

Run

init_bench

OR:

Run from the command line.

vectordbbench [OPTIONS] COMMAND [ARGS]...

To list the clients that are runnable via the commandline option, execute: vectordbbench --help

$ vectordbbench --help
Usage: vectordbbench [OPTIONS] COMMAND [ARGS]...
Options:
--help Show this message and exit.
Commands:
pgvectorhnsw
pgvectorivfflat
vectorchordrq
volcmysqlhnsw
test
weaviate

To list the options for each command, execute vectordbbench [command] --help

Use --note or --note-file to preserve deployment, resource, client, network, and constraint context in each result JSON under task_config.db_config.note. The options are mutually exclusive. Prefer --note-file for structured or multiline context, and never include credentials, tokens, or sensitive connection details.

vectordbbench zillizautoindex \
--note-file ./run-context.json \
<other options>

$ vectordbbench pgvectorhnsw --help
Usage: vectordbbench pgvectorhnsw [OPTIONS]
Options:
--config-file PATH Read configuration from yaml file
--drop-old / --skip-drop-old Drop old or skip [default: drop-old]
--load / --skip-load Load or skip [default: load]
--search-serial / --skip-search-serial
Search serial or skip [default: search-
serial]
--search-concurrent / --skip-search-concurrent
Search concurrent or skip [default: search-
concurrent]
--case-type [CapacityDim128|CapacityDim960|Performance768D100M|Performance768D10M|Performance768D1M|Performance768D10M1P|Performance768D1M1P|Performance768D10M99P|Performance768D1M99P|Performance1536D500K|Performance1536D5M|Performance1536D500K1P|Performance1536D5M1P|Performance1536D500K99P|Performance1536D5M99P|Performance1536D50K]
Case type
--db-label TEXT Db label, default: date in ISO format
[default: 2024-05-20T20:26:31.113290]
--note TEXT Run context stored with each result
[default: ""]
--note-file FILE Read run context from a UTF-8 text file
--dry-run Print just the configuration and exit
without running the tasks
--k INTEGER K value for number of nearest neighbors to
search [default: 100]
--concurrency-duration INTEGER Adjusts the duration in seconds of each
concurrency search [default: 30]
--num-concurrency TEXT Comma-separated list of concurrency values
to test during concurrent search [default:
1,10,20]
--concurrency-timeout INTEGER Timeout (in seconds) to wait for a
concurrency slot before failing. Set to a
negative value to wait indefinitely.
[default: 3600]
--user-name TEXT Db username [required]
--password TEXT Db password [required]
--host TEXT Db host [required]
--db-name TEXT Db name [required]
--maintenance-work-mem TEXT Sets the maximum memory to be used for
maintenance operations (index creation). Can
be entered as string with unit like '64GB'
or as an integer number of KB.This will set
the parameters:
max_parallel_maintenance_workers,
max_parallel_workers &
table(parallel_workers)
--max-parallel-workers INTEGER Sets the maximum number of parallel
processes per maintenance operation (index
creation)
--m INTEGER hnsw m
--ef-construction INTEGER hnsw ef-construction
--ef-search INTEGER hnsw ef-search
--quantization-type [none|bit|halfvec]
quantization type for vectors (in index)
--table-quantization-type [none|bit|halfvec]
quantization type for vectors (in table). If
equal to bit, the parameter
quantization_type will be set to bit too.
--reranking / --skip-reranking Enable reranking for HNSW search for binary
quantization
--reranking-metric [L2|COSINE|IP|DP]
Distance metric for reranking [default:
COSINE]
--quantized-fetch-limit INTEGER
Limit of fetching quantized vector ranked by
distance for reranking --
bound by ef_search
--custom-case-name TEXT Custom case name i.e. PerformanceCase1536D50K
--custom-case-description TEXT Custom name description
--custom-case-load-timeout INTEGER
Custom case load timeout [default: 36000]
--custom-case-optimize-timeout INTEGER
Custom case optimize timeout [default: 36000]
--custom-dataset-name TEXT
Dataset name i.e OpenAI
--custom-dataset-dir TEXT Dataset directory i.e. openai_medium_500k
--custom-dataset-size INTEGER Dataset size i.e. 500000
--custom-dataset-dim INTEGER Dataset dimension
--custom-dataset-metric-type TEXT
Dataset distance metric [default: COSINE]
--custom-dataset-file-count INTEGER
Dataset file count
--custom-dataset-use-shuffled / --skip-custom-dataset-use-shuffled
Use shuffled custom dataset or skip [default: custom-dataset-
use-shuffled]
--custom-dataset-with-gt / --skip-custom-dataset-with-gt
Custom dataset with ground truth or skip [default: custom-dataset-
with-gt]
--help Show this message and exit.

Run VectorChord (vchordrq) from command line

VectorChord is a PostgreSQL extension for scalable vector similarity search using IVF + RaBitQ indexing. It is fully compatible with pgvector data types and provides faster queries and index builds.

vectordbbench vectorchordrq \
--user-name postgres --password '<password>' \
--host localhost --port 5432 --db-name vectordb \
--case-type Performance1536D50K \
--lists 1000 --probes 10 --epsilon 1.9 \
--spherical-centroids --build-threads 8 \
--max-parallel-workers 15

Key VectorChord-specific options:

Option	Description
--lists	Number of IVF lists for vchordrq index
--probes	Number of probes during search (default: 10)
--epsilon	Reranking precision factor, 0.0-4.0 (default: 1.9)
--residual-quantization	Enable residual quantization
--spherical-centroids	L2-normalize centroids (recommended for cosine/IP)
--build-threads	Number of threads for index building (1-255)
--degree-of-parallelism	Degree of parallelism for index build (1-256)
--max-parallel-workers	Sets max_parallel_workers & max_parallel_maintenance_workers
--max-scan-tuples	Max tuples to scan before stopping (-1 for unlimited)

Run awsopensearch from command line

vectordbbench awsopensearch --db-label awsopensearch \
--m 16 --ef-construction 256 \
--host search-vector-db-prod-h4f6m4of6x7yp2rz7gdmots7w4.us-west-2.es.amazonaws.com --port 443 \
--user vector --password '<password>' \
--case-type Performance1536D5M --number-of-indexing-clients 10 \
--skip-load --num-concurrency 75

To list the options for awsopensearch, execute vectordbbench awsopensearch --help

$ vectordbbench awsopensearch --help
Usage: vectordbbench awsopensearch [OPTIONS]
Options:
# Sharding and Replication
--number-of-shards INTEGER Number of primary shards for the index
--number-of-replicas INTEGER Number of replica copies for each primary
shard
# Indexing Performance
--index-thread-qty INTEGER Thread count for native engine indexing
--index-thread-qty-during-force-merge INTEGER
Thread count during force merge operations
--number-of-indexing-clients INTEGER
Number of concurrent indexing clients
# Index Management
--number-of-segments INTEGER Target number of segments after merging
--refresh-interval TEXT How often to make new data available for
search
--force-merge-enabled BOOLEAN Whether to perform force merge operation
--flush-threshold-size TEXT Size threshold for flushing the transaction
log
--engine TEXT type of engine to use valid values [faiss, lucene, s3vector]
# Memory Management
--cb-threshold TEXT k-NN Memory circuit breaker threshold
--ondisk Ondisk mode with binary quantization(32x compression)
--oversample-factor Controls the degree of oversampling applied to minority classes in imbalanced datasets to improve model performance by balancing class distributions.(default 1.0)
# Quantization Type
--quantization-type TEXT which type of quantization to use valid values [fp32, fp16, bq]
--help Show this message and exit.

Run awsopensearch serverless from command line

OpenSearch Serverless (AOSS) is a serverless deployment option for Amazon OpenSearch Service. VDBBench supports AOSS with the --serverless flag, which uses AWS SigV4 authentication and automatically skips unsupported operations (cluster settings, force merge, manual refresh, warmup API).

Prerequisites:

• AWS credentials configured (via ~/.aws/credentials, environment variables, or IAM role)
• Serverless dependencies installed (included in the opensearch extra): pip install 'vectordb-bench[opensearch]'. This installs opensearch-py, boto3, and requests-aws4auth.
• IAM identity policy allowing aoss:APIAccessAll on the collection
• AOSS Data Access Policy granting index/collection permissions to the IAM principal

Example: Run performance test on OpenSearch Serverless

NUM_PER_BATCH=100 vectordbbench awsopensearch --db-label aoss \
--serverless --aws-region us-east-1 \
--host <collection-id>.aoss.us-east-1.on.aws --port 443 \
--case-type Performance768D1M \
--m 16 --ef-construction 200 --ef-search 40 \
--number-of-shards 8 --number-of-replicas 0 \
--engine faiss --metric-type cosine \
--num-concurrency 80,100,120

OpenSearch Serverless-specific options:

Option	Description
--serverless	Enable OpenSearch Serverless mode (uses AWS SigV4 auth)
--aws-region	AWS region for the AOSS collection (default: us-east-1)
NUM_PER_BATCH	Number of vectors per Serverless bulk request (default: 100)

Notes:

• --user and --password are not needed for Serverless mode
• --engine is accepted but ignored internally (AOSS manages the engine)
• --force-merge-enabled, --refresh-interval, --flush-threshold-size, and --cb-threshold are ignored for Serverless
• Keep NUM_PER_BATCH small enough for the Serverless bulk API request limits

Run Elastic Cloud from command line

Elastic Cloud supports multiple index types: HNSW, HNSW_INT8, HNSW_INT4, and HNSW_BBQ.

Example: Run HNSW index test

vectordbbench elasticcloudhnsw --db-label elastic-cloud-test \
--cloud-id <your-cloud-id> --password '<your-password>' \
--m 16 --ef-construction 100 --num-candidates 100 \
--case-type Performance768D1M --number-of-shards 1 \
--number-of-replicas 0 --refresh-interval 30s

Example: Run HNSW_INT8 index test

vectordbbench elasticcloudhnswint8 --db-label elastic-cloud-int8 \
--cloud-id <your-cloud-id> --password '<your-password>' \
--m 16 --ef-construction 200 --num-candidates 200 \
--case-type Performance1536D50K --element-type float

Example: Run HNSW_INT4 index test

vectordbbench elasticcloudhnswint4 --db-label elastic-cloud-int4 \
--cloud-id <your-cloud-id> --password '<your-password>' \
--m 16 --ef-construction 200 --num-candidates 200 \
--case-type Performance768D10M --use-rescore --oversample-ratio 2.0

Example: Run HNSW_BBQ index test

vectordbbench elasticcloudhnswbbq --db-label elastic-cloud-bbq \
--cloud-id <your-cloud-id> --password '<your-password>' \
--m 16 --ef-construction 200 --num-candidates 200 \
--case-type Performance1536D5M --use-routing --use-force-merge

Example: Run Label Filter Performance test

vectordbbench elasticcloudhnsw --db-label elastic-cloud-label-filter \
--cloud-id <your-cloud-id> --password '<your-password>' \
--case-type LabelFilterPerformanceCase \
--dataset-with-size-type "Medium OpenAI (1536dim, 500K)" \
--label-percentage 0.001 \
--m 16 --ef-construction 128 --num-candidates 100 \
--num-concurrency 1,5 --number-of-shards 1

To list all options for Elastic Cloud, execute vectordbbench elasticcloudhnsw --help. The following are Elastic Cloud-specific command-line options:

$ vectordbbench elasticcloudhnsw --help
Usage: vectordbbench elasticcloudhnsw [OPTIONS]
Options:
# Connection
--cloud-id TEXT Elastic Cloud ID [required]
--password TEXT Elastic Cloud password [required]
# HNSW Index Parameters
--m INTEGER HNSW M parameter [default: 16]
--ef-construction INTEGER HNSW efConstruction parameter [default: 100]
--num-candidates INTEGER Number of candidates for search [default: 100]
--element-type [float|byte] Element type for vectors (float: 4 bytes, byte: 1 byte) [default: float]
# Index Configuration
--number-of-shards INTEGER Number of shards [default: 1]
--number-of-replicas INTEGER Number of replicas [default: 0]
--refresh-interval TEXT Index refresh interval [default: 30s]
--merge-max-thread-count INTEGER
Maximum thread count for merge [default: 8]
--use-force-merge BOOLEAN Whether to use force merge [default: True]
--use-routing BOOLEAN Whether to use routing [default: False]
--use-rescore BOOLEAN Whether to use rescore [default: False]
--oversample-ratio FLOAT Oversample ratio for rescore [default: 2.0]
# Common Options
--case-type [CapacityDim128|CapacityDim960|Performance768D100M|...]
Case type
--db-label TEXT Db label, default: date in ISO format
--k INTEGER K value for number of nearest neighbors to search [default: 100]
--num-concurrency TEXT Comma-separated list of concurrency values [default: 1,5,10,20,30,40,60,80]
--help Show this message and exit.

Run OceanBase from command line

Execute tests for the index types: HNSW, HNSW_SQ, or HNSW_BQ.

vectordbbench oceanbasehnsw --host xxx --port xxx --user root@mysql_tenant --database test \
--m 16 --ef-construction 200 --case-type Performance1536D50K \
--index-type HNSW --ef-search 100

To list the options for oceanbase, execute vectordbbench oceanbasehnsw --help, The following are some OceanBase-specific command-line options.

$ vectordbbench oceanbasehnsw --help
Usage: vectordbbench oceanbasehnsw [OPTIONS]
Options:
[...]
--host TEXT OceanBase host
--user TEXT OceanBase username [required]
--password TEXT OceanBase database password
--database TEXT DataBase name [required]
--port INTEGER OceanBase port [required]
--m INTEGER hnsw m [required]
--ef-construction INTEGER hnsw ef-construction [required]
--ef-search INTEGER hnsw ef-search [required]
--index-type [HNSW|HNSW_SQ|HNSW_BQ]
Type of index to use. Supported values:
HNSW, HNSW_SQ, HNSW_BQ [required]
--help Show this message and exit.

Execute tests for the index types: IVF_FLAT, IVF_SQ8, or IVF_PQ.

vectordbbench oceanbaseivf --host xxx --port xxx --user root@mysql_tenant --database test \
--nlist 1000 --sample_per_nlist 256 --case-type Performance768D1M \
--index-type IVF_FLAT --ivf_nprobes 100

To list the options for oceanbase, execute vectordbbench oceanbaseivf --help, The following are some OceanBase-specific command-line options.

$ vectordbbench oceanbaseivf --help
Usage: vectordbbench oceanbaseivf [OPTIONS]
Options:
[...]
--host TEXT OceanBase host
--user TEXT OceanBase username [required]
--password TEXT OceanBase database password
--database TEXT DataBase name [required]
--port INTEGER OceanBase port [required]
--index-type [IVF_FLAT|IVF_SQ8|IVF_PQ]
Type of index to use. Supported values:
IVF_FLAT, IVF_SQ8, IVF_PQ [required]
--nlist INTEGER Number of cluster centers [required]
--sample_per_nlist INTEGER The cluster centers are calculated by total
sampling sample_per_nlist * nlist vectors
[required]
--ivf_nprobes TEXT How many clustering centers to search during
the query [required]
--m INTEGER The number of sub-vectors that each data
vector is divided into during IVF-PQ
--help Show this message and exit. Show this message and exit.

Run Hologres from command line

It is recommended to use the following code for installation.

pip install 'vectordb-bench[hologres]''psycopg[binary]' pgvector

Execute tests for the index types: HGraph.

NUM_PER_BATCH=10000 vectordbbench hologreshgraph --host Hologres_Endpoint --port 80 \
--user ACCESS_ID --password ACCESS_KEY --database DATABASE_NAME \
--m 64 --ef-construction 400 --case-type Performance768D10M \
--index-type HGraph --ef-search 400 --k 10 --num-concurrency 1,60,70,75,80,90,95,100,105,110,115,120,125,130 \
--serial-cooldown 3

To list the options for Hologres, execute vectordbbench hologreshgraph --help, The following are some Hologres-specific command-line options.

$ vectordbbench hologreshgraph --help
Usage: vectordbbench hologreshgraph [OPTIONS]
Options:
[...]
--host TEXT Hologres host
--user TEXT Hologres username [required]
--password TEXT Hologres database password
--database TEXT Hologres database name [required]
--port INTEGER Hologres port [required]
--m INTEGER hnsw m [required]
--ef-construction INTEGER hnsw ef-construction [required]
--ef-search INTEGER hnsw ef-search [required]
--index-type [HGraph] Type of index to use. Supported values:
HGraph [required]
--help Show this message and exit.

Run Zvec from command line

vectordbbench zvec --path Performance768D10M --db-label 16c64g-v0.1 \
--case-type Performance768D10M --num-concurrency 12,14,16,18,20 \
--quantize-type int8 --ef-search 118 --is-using-refiner

To list the options for zvec, execute vectordbbench zvec --help

 --path TEXT collection path [required]
--m INTEGER HNSW index parameter m.
--ef-construction INTEGER HNSW index parameter ef_construction
--ef-search INTEGER HNSW index parameter ef for search
--quantize-type TEXT HNSW index quantize type, fp16/int8
supported
--is-using-refiner is using refiner, suitable for quantized
index, recall `ef-search` results then
refine with unquantized vector to `topk`
results

Run Doris from command line

Doris supports ann index with type hnsw from version 4.0.x

NUM_PER_BATCH=1000000 vectordbbench doris --http-port=8030 --port=9030 --db-name=vector_test --case-type=Performance768D1M --stream-load-rows-per-batch=500000

Using flag --session-var, if you want to test doris with some customized session variables. For example:

NUM_PER_BATCH=1000000 vectordbbench doris --http-port=8030 --port=9030 --db-name=vector_test --case-type=Performance768D1M --stream-load-rows-per-batch=500000 --session-var enable_profile=True

Mote options:

--m INTEGER hnsw m
--ef-construction INTEGER hnsw ef-construction
--username TEXT Username [default: root; required]
--password TEXT Password [default: ""]
--host TEXT Db host [default: 127.0.0.1; required]
--port INTEGER Query Port [default: 9030; required]
--http-port INTEGER Http Port [default: 8030; required]
--db-name TEXT Db name [default: test; required]
--ssl / --no-ssl Enable or disable SSL, for Doris Serverless
SSL must be enabled [default: no-ssl]
--index-prop TEXT Extra index PROPERTY as key=value
(repeatable)
--session-var TEXT Session variable key=value applied to each
SQL session (repeatable)
--stream-load-rows-per-batch INTEGER
Rows per single stream load request; default
uses NUM_PER_BATCH
--no-index Create table without ANN index

Run Lindorm from command line

Lindorm supports index types: hnsw, ivfpq, or ivfbq.

Example: Run hnsw index test

vectordbbench lindormhnsw --case-type Performance768D10M --index-name <index_name> --k 10 \
--host <lindorm_host> --port <lindorm_port> --user <username> --password <password> --m 32 \
--ef-construction 400 --ef-search 150 --reorder-factor 2

Example: Run ivfpq index test

vectordbbench lindormivfpq --case-type Performance768D10M \
--index-name <index_name> --k 10 --host <lindorm_host> --port <lindorm_port> \
--user <username> --password <password> --lists <nlist> --probes <nprobe> \
--m 32 --ef-construction 500 --ef-search 200 --reorder-factor 2

Example: Run ivfbq index test

vectordbbench lindormivfbq --case-type Performance768D10M --index-name <index_name> \
--k 10 --host <index_name> --port <lindorm_port> \
--user <username> --password <password> --lists <nlist> --probes <nprobe> \
--exbits 2 --m 32 --ef-construction 500 --ef-search 200 --reorder-factor 2

To list the options for Lindorm, execute vectordbbench lindormhnsw --help, The following are some Lindorm-specific command-line options.

 --host TEXT host connection string [required]
--port INTEGER Db Port [required]
--user TEXT Db username [required]
--password TEXT Db password [required]
--index-name TEXT Db index name [required]
--filter-type TEXT post_filter|pre_filter|efficient_filter
--number-of-regions INTEGER Vector number of regions
--m INTEGER hnsw m [required]
--ef-construction INTEGER hnsw ef-construction [required]
--ef-search INTEGER hnsw ef-search [required]
--reorder-factor INTEGER reorder factor

Run ADBPG (Aliyun AnalyticDB for PostgreSQL) from command line

ADBPG Nova uses the fastann/Nova vector index engine with USING ann syntax.

Example: Run novamr index benchmark (BioASQ 1M, 1024-dim)

vectordbbench adbpgnova --case-type Performance1024D1M --k 10 \
--host <adbpg_host> --port 5432 --db-name postgres \
--user-name <username> --password <password> \
--algorithm novamr --hnsw-m 48 --ef-construction 600 \
--ef-search 130 --max-scan-points 5000 --quantize-rescore-amp 2.0

Example: Run from config file

vectordbbench adbpgnova --config-file adbpg_bioasq1m_novamr.yml

To list the options for ADBPG, execute vectordbbench adbpgnova --help. The following are some ADBPG-specific command-line options.

 --user-name TEXT Db username [required]
--password TEXT Postgres database password [$POSTGRES_PASSWORD]
--host TEXT Db host [required]
--port INTEGER Postgres database port [default: 5432]
--db-name TEXT Db name [required]
--algorithm TEXT algorithm [default: novamr]
--hnsw-m INTEGER hnsw_m [default: 16]
--ef-construction INTEGER ef_construction [default: 200]
--ef-search INTEGER ef_search [default: 100]
--max-scan-points INTEGER max scan points [default: 2000]
--quantize-rescore-amp FLOAT fastann.quantize_rescore_amp [default: 1.0]
--nova-adaptive-gamma FLOAT fastann.nova_adaptive_gamma [default: 0.0]
--auto-reduction/--no-auto-reduction Index WITH auto_reduction=on [default: False]

Run PolarDB from command line

PolarDB supports index types: faiss_hnsw_flat, faiss_hnsw_pq, and faiss_hnsw_sq.

Example: Run faiss_hnsw_flat benchmark

vectordbbench polardbhnswflat \
--case-type Performance768D1M \
--username <db_user> \
--password '<db_password>' \
--host <db_host> \
--port 3306 \
--m 16 \
--ef-construction 256 \
--ef-search 256 \
--insert-workers 64 \
--num-concurrency '10,20,40,60,80' \
--concurrency-duration 60 \
--task-label <task_label> \
--db-label <db_label> \
--skip-search-serial \
--post-load-index

To list the options for PolarDB, execute vectordbbench polardbhnswflat --help. The following are some PolarDB-specific command-line options.

 --username TEXT Username [required]
--password TEXT Password
--host TEXT Db host [default: 127.0.0.1]
--port INTEGER Db Port [default: 3306]
--database TEXT Database name [default: vectordbbench]
--m INTEGER M parameter (max_degree) in HNSW
--ef-construction INTEGER ef_construction parameter in HNSW
--ef-search INTEGER polar_vector_index_hnsw_ef_search session variable
--insert-workers INTEGER Number of concurrent threads for data insertion
--post-load-index / --inline-index
Create index after load or inline at table creation

Run VolcMySQL from command line

VolcMySQL is a MySQL-compatible service with a native VECTOR type and an HNSW vector index (created via SECONDARY_ENGINE_ATTRIBUTE). Optional quantization is configurable through --quant-algorithm (NONE, SQ, PQ) and --quant-type (16_bit, 8_bit, 4_bit, binary).

vectordbbench volcmysqlhnsw \
--case-type Performance1536D50K \
--username <db_user> \
--password '<db_password>' \
--host <db_host> \
--port 3306 \
--m 16 \
--ef-construction 128 \
--ef-search 100 \
--quant-algorithm SQ \
--quant-type 16_bit \
--num-concurrency '10,20,40,60,80' \
--concurrency-duration 30 \
--task-label <task_label> \
--db-label <db_label>

To list the options for VolcMySQL, execute vectordbbench volcmysqlhnsw --help. The following are some VolcMySQL-specific command-line options.

 --username TEXT Username [required]
--password TEXT Password [required]
--host TEXT Db host [default: 127.0.0.1]
--port INTEGER DB Port [default: 3306]
--m INTEGER M parameter in HNSW vector indexing
--ef-search INTEGER Session variable loose_hnsw_ef_search
--ef-construction INTEGER HNSW ef_construction
--quant-algorithm [NONE|SQ|PQ] Quantization algorithm
--quant-type [16_bit|8_bit|4_bit|binary]
Quantization type

Note: vectors are loaded and queried over the raw-binary VECTOR path by default; the client auto-probes server support and falls back to to_vector() text when unavailable. Set VDB_BINARY_VEC=0 to force the text path or 1 to force binary.

Using a configuration file.

The vectordbbench command can optionally read some or all the options from a yaml formatted configuration file.

By default, configuration files are expected to be in vectordb_bench/config-files/, this can be overridden by setting the environment variable CONFIG_LOCAL_DIR or by passing the full path to the file.

The required format is:

commandname:
parameter_name: parameter_valueparameter_name: parameter_value

Example:

pgvectorhnsw:
db_label: pgConfigTestuser_name: vectordbbenchpassword: vectordbbenchdb_name: vectordbbenchhost: localhostm: 16ef_construction: 128ef_search: 128milvushnsw:
skip_search_serial: Truecase_type: Performance1536D50Kuri: http://localhost:19530m: 16ef_construction: 128ef_search: 128drop_old: Falseload: Falseelasticcloudhnsw:
db_label: elastic-cloud-hnswcloud_id: <your-cloud-id>password: <your-password>case_type: Performance768D1Mm: 16ef_construction: 100num_candidates: 100number_of_shards: 1number_of_replicas: 0refresh_interval: 30selement_type: float

Notes:

• Options passed on the command line will override the configuration file*
• Parameter names use an _ not -
• For LabelFilterPerformanceCase and NewIntFilterPerformanceCase, you must specify dataset_with_size_type in addition to case_type

Using a batch configuration file.

The vectordbbench command can read a batch configuration file to run all the test cases in the yaml formatted configuration file.

By default, configuration files are expected to be in vectordb_bench/config-files/, this can be overridden by setting the environment variable CONFIG_LOCAL_DIR or by passing the full path to the file.

The required format is:

commandname:
- parameter_name: parameter_valueanother_parameter_name: parameter_value

Example:

pgvectorhnsw:
- db_label: pgConfigTestuser_name: vectordbbenchpassword: vectordbbenchdb_name: vectordbbenchhost: localhostm: 16ef_construction: 128ef_search: 128milvushnsw:
- skip_search_serial: Truecase_type: Performance1536D50Kuri: http://localhost:19530m: 16ef_construction: 128ef_search: 128drop_old: Falseload: Falseelasticcloudhnsw:
- db_label: elastic-cloud-hnsw-test-1cloud_id: <your-cloud-id>password: <your-password>case_type: Performance768D1Mm: 16ef_construction: 100num_candidates: 100
- db_label: elastic-cloud-label-filter-0.1cloud_id: <your-cloud-id>password: <your-password>case_type: LabelFilterPerformanceCasedataset_with_size_type: "Medium OpenAI (1536dim, 500K)"label_percentage: 0.001m: 16ef_construction: 128num_candidates: 100num_concurrency: "1,5"

Notes:

• Options can only be passed through configuration files
• Parameter names use an _ not -
• For LabelFilterPerformanceCase and NewIntFilterPerformanceCase, you must specify dataset_with_size_type in addition to case_type

How to use?

vectordbbench batchcli --batch-config-file <your-yaml-configuration-file>

Leaderboard

Introduction

To facilitate the presentation of test results and provide a comprehensive performance analysis report, we offer a leaderboard page. It allows us to choose from QPS, QP$, and latency metrics, and provides a comprehensive assessment of a system's performance based on the test results of various cases and a set of scoring mechanisms (to be introduced later). On this leaderboard, we can select the systems and models to be compared, and filter out cases we do not want to consider. Comprehensive scores are always ranked from best to worst, and the specific test results of each query will be presented in the list below.

Cloud Leaderboard

VectorDBBench now includes Cloud Leaderboard cases for production-oriented cloud vector database evaluation. These cases complement the original raw-performance leaderboard by measuring behaviors that matter for managed services:

• CloudInsertCase: insert throughput plus searchable and indexed readiness delays.
• CloudPayloadSearchCase: search performance when responses return IDs only, scalar metadata, or vectors.
• CloudMultiTenantSearchCase: tenant-routed search for SaaS-shaped workloads.
• CloudColdLatencyCase: cold and warm serial latency for first-query and cache-sensitive serving paths.

The May 2026 release note explains why the Cloud Leaderboard was added, what changed, which systems were tested this round, and how to run each new case: docs/release/2026-05-cloud-leaderboard.md.

Scoring Rules

1. For each case, select a base value and score each system based on relative values.

   • For QPS and QP$, we use the highest value as the reference, denoted as base_QPS or base_QP$, and the score of each system is (QPS/base_QPS) * 100 or (QP$/base_QP$) * 100.
   • For Latency, we use the lowest value as the reference, that is, base_Latency, and the score of each system is (base_Latency + 10ms)/(Latency + 10ms) * 100.

   We want to give equal weight to different cases, and not let a case with high absolute result values become the sole reason for the overall scoring. Therefore, when scoring different systems in each case, we need to use relative values.

   Also, for Latency, we add 10ms to the numerator and denominator to ensure that if every system performs particularly well in a case, its advantage will not be infinitely magnified when latency tends to 0.

2. For systems that fail or timeout in a particular case, we will give them a score based on a value worse than the worst result by a factor of two. For example, in QPS or QP$, it would be half the lowest value. For Latency, it would be twice the maximum value.

3. For each system, we will take the geometric mean of its scores in all cases as its comprehensive score for a particular metric.

Build on your own

Install requirements

pip install -e '.[test]'
pip install -e '.[pinecone]'

Run test server

python -m vectordb_bench

OR:

init_bench

OR:

If you are using dev container, create the following dataset directory first:

# Mount local ~/vectordb_bench/dataset to contain's /tmp/vectordb_bench/dataset.# If you are not comfortable with the path name, feel free to change it in devcontainer.json
mkdir -p ~/vectordb_bench/dataset

After reopen the repository in container, run python -m vectordb_bench in the container's bash.

Check coding styles

make lint

To fix the coding styles automatically

make format

How does it work?

Result Page

This is the main page of VDBBench, which displays the standard benchmark results we provide. Additionally, results of all tests performed by users themselves will also be shown here. We also offer the ability to select and compare results from multiple tests simultaneously.

The standard benchmark results displayed here include all 15 cases that we currently support for 6 of our clients (Milvus, Zilliz Cloud, Elastic Search, Qdrant Cloud, Weaviate Cloud and PgVector). However, as some systems may not be able to complete all the tests successfully due to issues like Out of Memory (OOM) or timeouts, not all clients are included in every case.

All standard benchmark results are generated by a client running on an 8 core, 32 GB host, which is located in the same region as the server being tested. The client host is equipped with an Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz processor. Also all the servers for the open-source systems tested in our benchmarks run on hosts with the same type of processor.

Run Test Page

1. Initially, you select the systems to be tested - multiple selections are allowed. Once selected, corresponding forms will pop up to gather necessary information for using the chosen databases. The db_label is used to differentiate different instances of the same system. We recommend filling in the host size or instance type here (as we do in our standard results).
2. The next step is to select the test cases you want to perform. You can select multiple cases at once, and a form to collect corresponding parameters will appear.
3. Finally, you'll need to provide a task label to distinguish different test results. Using the same label for different tests will result in the previous results being overwritten. Now we can only run one task at the same time.

Module

Code Structure

Client

Our client module is designed with flexibility and extensibility in mind, aiming to integrate APIs from different systems seamlessly. As of now, it supports Milvus, Zilliz Cloud, Elastic Search, Pinecone, Qdrant Cloud, Weaviate Cloud, PgVector, VectorChord, Redis, Chroma, CockroachDB, VolcMySQL, etc. Stay tuned for more options, as we are consistently working on extending our reach to other systems.

Benchmark Cases

We've developed lots of comprehensive benchmark cases to test vector databases' various capabilities, each designed to give you a different piece of the puzzle. These cases are categorized into several main types:

Capacity Case

• Large Dim: Tests the database's loading capacity by inserting large-dimension vectors (GIST 100K vectors, 960 dimensions) until fully loaded. The final number of inserted vectors is reported.
• Small Dim: Similar to the Large Dim case but uses small-dimension vectors (SIFT 500K vectors, 128 dimensions).

Search Performance Case

• XLarge Dataset: Measures search performance with a massive dataset (LAION 100M vectors, 768 dimensions) at varying parallel levels. The results include index building time, recall, latency, and maximum QPS.
• Large Dataset: Similar to the XLarge Dataset case, but uses a slightly smaller dataset (10M-1024dim, 10M-768dim, 5M-1536dim).
• Medium Dataset: A case using a medium dataset (1M-1024dim, 1M-768dim, 500K-1536dim).
• Small Dataset: For development (100K-768dim, 50K-1536dim).

Filtering Search Performance Case

• Int-Filter Cases: Evaluates search performance with int-based filter expression (e.g. "id >= 2,000").
• Label-Filter Cases: Evaluates search performance with label-based filter expressions (e.g., "color == 'red'"). The test includes randomly generated labels to simulate real-world filtering scenarios.

Full Text Search Performance Case

• FullTextSearchPerformance: Measures BM25-style text retrieval over raw text documents. The case inserts documents, runs the backend optimization or index-readiness step, then measures recall, latency, and QPS for text queries.
• Datasets: The initial FTS benchmark uses MS MARCO and HotpotQA in small, medium, and large corpus sizes.
• Ground truth: Recall, MRR, and NDCG are computed against positive semantic relevance labels from ir_datasets.
• Payload profiles: FTS supports IDs-only responses and text payload responses so users can compare pure retrieval throughput against response-size overhead.

Streaming Cases

• Insertion-Under-Load Case: Evaluates search performance while maintaining a constant insertion workload. VDBBench applies a steady stream of insert requests at a fixed rate to simulate real-world scenarios where search operations must perform reliably under continuous data ingestion.

Each case provides an in-depth examination of a vector database's abilities, providing you a comprehensive view of the database's performance.

Custom Dataset for Performance case

Through the /custom page, users can customize their own performance case using local datasets. After saving, the corresponding case can be selected from the /run_test page to perform the test.

We have strict requirements for the data set format, please follow them.

• Folder Path - The path to the folder containing all the files. Please ensure that all files in the folder are in the Parquet format.

  • Vectors data files: The file must be named train.parquet and should have two columns: id as an incrementing int and emb as an array of float32.
  • Query test vectors: The file must be named test.parquet and should have two columns: id as an incrementing int and emb as an array of float32.
    • We recommend limiting the number of test query vectors, like 1,000. When conducting concurrent query tests, Vdbbench creates a large number of processes. To minimize additional communication overhead during testing, we prepare a complete set of test queries for each process, allowing them to run independently. However, this means that as the number of concurrent processes increases, the number of copied query vectors also increases significantly, which can place substantial pressure on memory resources.
  • Ground truth file: The file must be named neighbors.parquet and should have two columns: id corresponding to query vectors and neighbors_id as an array of int.

• Train File Count - If the vector file is too large, you can consider splitting it into multiple files. The naming format for the split files should be train-[index]-of-[file_count].parquet. For example, train-01-of-10.parquet represents the second file (0-indexed) among 10 split files.

• Use Shuffled Data - If you check this option, the vector data files need to be modified. VDBBench will load the data labeled with shuffle. For example, use shuffle_train.parquet instead of train.parquet and shuffle_train-04-of-10.parquet instead of train-04-of-10.parquet. The id column in the shuffled data can be in any order.

Goals

Our goals of this benchmark are:

Reproducibility & Usability

One of the primary goals of VDBBench is to enable users to reproduce benchmark results swiftly and easily, or to test their customized scenarios. We believe that lowering the barriers to entry for conducting these tests will enhance the community's understanding and improvement of vector databases. We aim to create an environment where any user, regardless of their technical expertise, can quickly set up and run benchmarks, and view and analyze results in an intuitive manner.

Representability & Realism

VDBBench aims to provide a more comprehensive, multi-faceted testing environment that accurately represents the complexity of vector databases. By moving beyond a simple speed test for algorithms, we hope to contribute to a better understanding of vector databases in real-world scenarios. By incorporating as many complex scenarios as possible, including a variety of test cases and datasets, we aim to reflect realistic conditions and offer tangible significance to our community. Our goal is to deliver benchmarking results that can drive tangible improvements in the development and usage of vector databases.

Contribution

General Guidelines

1. Fork the repository and create a new branch for your changes.
2. Adhere to coding conventions and formatting guidelines.
3. Use clear commit messages to document the purpose of your changes.

Adding New Clients

Step 1: Creating New Client Files

1. Navigate to the vectordb_bench/backend/clients directory.
2. Create a new folder for your client, for example, "new_client".
3. Inside the "new_client" folder, create two files: new_client.py and config.py.

Step 2: Implement new_client.py and config.py

1. Open new_client.py and define the NewClient class, which should inherit from the clients/api.py file's VectorDB abstract class. The VectorDB class serves as the API for benchmarking, and all DB clients must implement this abstract class. Example implementation in new_client.py: new_client.py

from ..apiimportVectorDBclassNewClient(VectorDB):
# Implement the abstract methods defined in the VectorDB class# ...

2. Open config.py and implement the DBConfig and optional DBCaseConfig classes.
3. The DBConfig class should be an abstract class that provides information necessary to establish connections with the database. It is recommended to use the pydantic.SecretStr data type to handle sensitive data such as tokens, URIs, or passwords.
4. The DBCaseConfig class is optional and allows for providing case-specific configurations for the database. If not provided, it defaults to EmptyDBCaseConfig. Example implementation in config.py:

frompydanticimportSecretStrfromclients.apiimportDBConfig, DBCaseConfigclassNewDBConfig(DBConfig):
# Implement the required configuration fields for the database connection# ...token: SecretStruri: strclassNewDBCaseConfig(DBCaseConfig):
# Implement optional case-specific configuration fields# ...

Step 3: Importing the DB Client and Updating Initialization

In this final step, you will import your DB client into clients/init.py and update the initialization process.

1. Open clients/init.py and import your NewClient from new_client.py.
2. Add your NewClient to the DB enum.
3. Update the db2client dictionary by adding an entry for your NewClient. Example implementation in clients/init.py:

#clients/__init__.py# Add NewClient to the DB enumclassDB(Enum):
...
DB.NewClient="NewClient"@propertydefinit_cls(self) ->Type[VectorDB]:
...
ifself==DB.NewClient:
from .new_client.new_clientimportNewClientreturnNewClient
...
@propertydefconfig_cls(self) ->Type[DBConfig]:
...
ifself==DB.NewClient:
from .new_client.configimportNewClientConfigreturnNewClientConfig
...
defcase_config_cls(self, ...)
ifself==DB.NewClient:
from .new_client.configimportNewClientCaseConfigreturnNewClientCaseConfig

Step 4: Implement new_client/cli.py and vectordb_bench/cli/vectordbbench.py

In this (optional, but encouraged) step you will enable the test to be run from the command line.

1. Navigate to the vectordb_bench/backend/clients/"client" directory.
2. Inside the "client" folder, create a cli.py file. Using zilliz as an example cli.py:

fromtypingimportAnnotated, UnpackimportclickimportosfrompydanticimportSecretStrfromvectordb_bench.cli.cliimport (
CommonTypedDict,
cli,
click_parameter_decorators_from_typed_dict,
run,
)
fromvectordb_bench.backend.clientsimportDBclassZillizTypedDict(CommonTypedDict):
uri: Annotated[
str, click.option("--uri", type=str, help="uri connection string", required=True)
]
user_name: Annotated[
str, click.option("--user-name", type=str, help="Db username", required=True)
]
password: Annotated[
str,
click.option("--password",
type=str,
help="Zilliz password",
default=lambda: os.environ.get("ZILLIZ_PASSWORD", ""),
show_default="$ZILLIZ_PASSWORD",
),
]
level: Annotated[
str,
click.option("--level", type=str, help="Zilliz index level", required=False),
]
@cli.command()@click_parameter_decorators_from_typed_dict(ZillizTypedDict)defZillizAutoIndex(**parameters: Unpack[ZillizTypedDict]):
from .configimportZillizCloudConfig, AutoIndexConfigrun(
db=DB.ZillizCloud,
db_config=ZillizCloudConfig(
db_label=parameters["db_label"],
uri=SecretStr(parameters["uri"]),
user=parameters["user_name"],
password=SecretStr(parameters["password"]),
),
db_case_config=AutoIndexConfig(
params={parameters["level"]},
),
**parameters,
)

3. Update cli by adding:
   1. Add database specific options as an Annotated TypedDict, see ZillizTypedDict above.
   2. Add index configuration specific options as an Annotated TypedDict. (example: vectordb_bench/backend/clients/pgvector/cli.py)
      1. May not be needed if there is only one index config.
      2. Repeat for each index configuration, nesting them if possible.
   3. Add a index config specific function for each index type, see Zilliz above. The function name, in lowercase, will be the command name passed to the vectordbbench command.
   4. Update db_config and db_case_config to match client requirements
   5. Continue to add new functions for each index config.
   6. Import the client cli module and command to vectordb_bench/cli/vectordbbench.py (for databases with multiple commands (index configs), this only needs to be done for one command)
   7. Import the get_custom_case_config function from vectordb_bench/cli/cli.py and use it to add a new key custom_case to the parameters variable within the command.

cli modules with multiple index configs:

• pgvector: vectordb_bench/backend/clients/pgvector/cli.py
• milvus: vectordb_bench/backend/clients/milvus/cli.py

That's it! You have successfully added a new DB client to the vectordb_bench project.

Rules

Installation

The system under test can be installed in any form to achieve optimal performance. This includes but is not limited to binary deployment, Docker, and cloud services.

Fine-Tuning

For the system under test, we use the default server-side configuration to maintain the authenticity and representativeness of our results. For the Client, we welcome any parameter tuning to obtain better results.

Incomplete Results

Many databases may not be able to complete all test cases due to issues such as Out of Memory (OOM), crashes, or timeouts. In these scenarios, we will clearly state these occurrences in the test results.

Mistake Or Misrepresentation

We strive for accuracy in learning and supporting various vector databases, yet there might be oversights or misapplications. For any such occurrences, feel free to raise an issue or make amendments on our GitHub page.

Timeout

In our pursuit to ensure that our benchmark reflects the reality of a production environment while guaranteeing the practicality of the system, we have implemented a timeout plan based on our experiences for various tests.

1. Capacity Case:

• For Capacity Case, we have assigned an overall timeout.

2. Other Cases:

For other cases, we have set two timeouts:

• Data Loading Timeout: This timeout is designed to filter out systems that are too slow in inserting data, thus ensuring that we are only considering systems that is able to cope with the demands of a real-world production environment within a reasonable time frame.

• Optimization Preparation Timeout: This timeout is established to avoid excessive optimization strategies that might work for benchmarks but fail to deliver in real production environments. By doing this, we ensure that the systems we consider are not only suitable for testing environments but also applicable and efficient in production scenarios.

This multi-tiered timeout approach allows our benchmark to be more representative of actual production environments and assists us in identifying systems that can truly perform in real-world scenarios.

Case	Data Size	Timeout Type	Value
Capacity Case	N/A	Loading timeout	24 hours
Other Cases	1M vectors, 768 dimensions 500K vectors, 1536 dimensions	Loading timeout	2.5 hours
		Optimization timeout	15 mins
Other Cases	10M vectors, 768 dimensions 5M vectors, 1536 dimensions	Loading timeout	25 hours
		Optimization timeout	2.5 hours
Other Cases	100M vectors, 768 dimensions	Loading timeout	250 hours
		Optimization timeout	25 hours

Note: Some datapoints in the standard benchmark results that violate this timeout will be kept for now for reference. We will remove them in the future.