Benchmarks.

Fast and accurate retrieval is the backbone of the new generation of AI-native applications.
TopK is built for teams working with large volumes of data where high performance, flexibility, and cost efficiency matter.

Enterprise retrieval

Find the details that dense embeddings miss.

Accuracy across real-world domains

Recall · higher is better

TopK Multi-VectorDense Embeddings (Qwen3 VLE 8B)

Page-level recall comparison across financial filings, industrial, pharmaceutical, and HR documents. TopK multi-vector uses topk-io/Iso-ModernColBERT model.

Dataset / model	TopK Multi-Vector	Dense Embeddings (Qwen3 VLE 8B)
Financial filings	80.9%	58.9%
Pharmaceutical	84%	60%
HR	81%	68%
Industrial	76%	42%

Agentic search

Better retrieval. Better agents.

BrowseComp-Plus

Accuracy and recall · higher is better

TopK Multi-VectorDense Embeddings (Qwen3 8B)BM25

BrowseComp-Plus with GPT-5 agent and the default retrieval/eval harness. Compares the TopK Semantic Index (multi-vector), Qwen3 Embedding 8B (dense), and BM25. Official result accepted by the benchmark leaderboard.

Metric	TopK Multi-Vector	Dense Embeddings (Qwen3 8B)	BM25
Accuracy	80.48%	71.69%	57.59%
Recall	77.82%	78.98%	61.7%

Hybrid search

Combine multiple signals to optimize relevance.

Ranking signals working together

nDCG@10 · higher is better

TopK HybridReciprocal Rank Fusion (RRF)

TopK hybrid scoring compared with reciprocal rank fusion (RRF), measured by nDCG@10 on NFCorpus, FiQA, FEVER, and Natural Questions (NQ).

Dataset / model	TopK Hybrid	Reciprocal Rank Fusion (RRF)
FEVER	0.865	0.843
NQ	0.552	0.541
FiQA	0.429	0.415
NFCorpus	0.368	0.350

Billion documents in milliseconds

Predictable latency at scale with filtered or unfiltered queries.

p50p99

LOWER IS BETTER

768d dense vectors. Hot cache.10M → 1B

View latency data

Documents	p50 · No filter	p99 · No filter
10M	15.5 ms	17 ms
100M	31.2 ms	34.8 ms
1B	52.9 ms	61.8 ms

Latency measured on a production-config cluster with randomly generated documents and queries.

Ship better
search today.

Build on flexible primitives and focus on
what makes your beer taste better.

Start building Talk to us

Read the docs