Nebius Customer AtlasNot affiliated · Research by Not affiliated · By @anthonybs
Not affiliated with Nebius · Independent research by @anthonybsSee changes
AllLinkup

Linkup

Web search APIs for AI agents

How they use Nebius and Tavily

Linkup runs its Quality Scorer model on Nebius GPUs to choose web pages for its search index. Co-founder and COO Boris also describes embedding workloads, batching, scaling and a reliable fallback route. His interview identifies the GPU deployment as Nebius Cloud and is presented by Nebius as a Token Factory customer account.

Linkup also used Tavily for a SimpleQA benchmark evaluation. It ran the same evaluation script used for its own search API because it could not find published Tavily results. Linkup’s repository identifies Tavily’s advanced search mode and reports an F-score of 0.728, accuracy of 0.726 and average latency of 2.92 seconds.

InfrastructureService: Nebius AI CloudService: Nebius Token FactoryService: TavilyInferenceData processingSearch & retrieval

Sources

3 sources

Nebius Linkup COO · quality scoring and embeddings

27 Sept 2026 · Nebius · Video

At 0:40, Boris says Linkup has run its Quality Scorer on Nebius GPUs for a while. At 1:14–2:49 he describes embedding scale, batching and fallback reliability. The original title identifies Token Factory.

Open original source at 0:40
Nebius Linkup COO · quality scoring and embeddings27 Sept 2026 · Nebius · Video · Starts at 0:40At 0:40, Boris says Linkup has run its Quality Scorer on Nebius GPUs for a while. At 1:14–2:49 he describes embedding scale, batching and fallback reliability. The original title identifies Token Factory.

eval-simpleQA/README.md at main · LinkupPlatform/eval-simpleQA · GitHub

14 Oct 2026 · github.com · Web page

Linkup reports evaluating Tavily’s advanced search API on OpenAI’s SimpleQA benchmark. It reports an F-score of 0.728, accuracy of 0.726, an attempt rate of 99.6% and average latency of 2.92 seconds.

Open original source
eval-simpleQA/README.md at main · LinkupPlatform/eval-simpleQA · GitHub14 Oct 2026 · github.com · Web pageLinkup reports evaluating Tavily’s advanced search API on OpenAI’s SimpleQA benchmark. It reports an F-score of 0.728, accuracy of 0.726, an attempt rate of 99.6% and average latency of 2.92 seconds.

Linkup - Linkup has achieved SOTA performance

14 Oct 2026 · www.linkup.so · Web page

Linkup states that it evaluated Tavily on SimpleQA using the same script used to measure its own search API. It reports a rounded Tavily score of 73%.

Open original source
Linkup - Linkup has achieved SOTA performance14 Oct 2026 · www.linkup.so · Web pageLinkup states that it evaluated Tavily on SimpleQA using the same script used to measure its own search API. It reports a rounded Tavily score of 73%.

Company context

linkup.so

Linkup builds web search and retrieval infrastructure for AI developers and enterprise teams. Its API connects AI agents and language models to current web information for research, legal work, sales enrichment and other applications. Its main products are Fetch, which extracts content from individual webpages; Search, which returns web results and source-cited answers; and Research, which handles more complex investigations. Enterprise options include dedicated indexes, access controls and deployment within customers’ own cloud infrastructure.

Documented financing · private company

Amount$10MSeed
· Linkup press release Read follow-up source

Recent updates

7 updates

  1. Partnership

    Linkup announces search partnership with Aleph Alpha

    Linkup announced that its web search API powers Aleph Alpha’s internal tools and customer-facing AI solutions. The integration runs on EU infrastructure with zero data retention and supports industrial, public-sector and defense customers.

    Linkup

  2. Partnership

    Linkup partners with Cloudflare for AI Gateway web search

    Cloudflare launched its Web Search API with Linkup among its initial search partners. Developers can access live web results through AI Gateway with unified billing, observability and access controls.

    Cloudflare Blog

  3. Open model

    Linkup releases SPARSEUP retrieval model

    Linkup Research published an open sparse embedding model and its benchmark discussion.

    Linkup research post
Next, A–ZLioma
← Explore all customers

Linkup

Web retrieval and agent search APIs · Infrastructure

How they use Nebius and Tavily

Linkup runs its Quality Scorer model on Nebius GPUs to choose web pages for its search index. Co-founder and COO Boris also describes embedding workloads, batching, scaling and a reliable fallback route. His interview identifies the GPU deployment as Nebius Cloud and is presented by Nebius as a Token Factory customer account.

Linkup also used Tavily for a SimpleQA benchmark evaluation. It ran the same evaluation script used for its own search API because it could not find published Tavily results. Linkup’s repository identifies Tavily’s advanced search mode and reports an F-score of 0.728, accuracy of 0.726 and average latency of 2.92 seconds.

InfrastructureService: Nebius AI CloudService: Nebius Token FactoryService: TavilyInferenceData processingSearch & retrieval

Sources

3 sources

Nebius Linkup COO · quality scoring and embeddings

27 Sept 2026 · Nebius · Video

At 0:40, Boris says Linkup has run its Quality Scorer on Nebius GPUs for a while. At 1:14–2:49 he describes embedding scale, batching and fallback reliability. The original title identifies Token Factory.

Open original source at 0:40
Nebius Linkup COO · quality scoring and embeddings27 Sept 2026 · Nebius · Video · Starts at 0:40At 0:40, Boris says Linkup has run its Quality Scorer on Nebius GPUs for a while. At 1:14–2:49 he describes embedding scale, batching and fallback reliability. The original title identifies Token Factory.

eval-simpleQA/README.md at main · LinkupPlatform/eval-simpleQA · GitHub

14 Oct 2026 · github.com · Web page

Linkup reports evaluating Tavily’s advanced search API on OpenAI’s SimpleQA benchmark. It reports an F-score of 0.728, accuracy of 0.726, an attempt rate of 99.6% and average latency of 2.92 seconds.

Open original source
eval-simpleQA/README.md at main · LinkupPlatform/eval-simpleQA · GitHub14 Oct 2026 · github.com · Web pageLinkup reports evaluating Tavily’s advanced search API on OpenAI’s SimpleQA benchmark. It reports an F-score of 0.728, accuracy of 0.726, an attempt rate of 99.6% and average latency of 2.92 seconds.

Linkup - Linkup has achieved SOTA performance

14 Oct 2026 · www.linkup.so · Web page

Linkup states that it evaluated Tavily on SimpleQA using the same script used to measure its own search API. It reports a rounded Tavily score of 73%.

Open original source
Linkup - Linkup has achieved SOTA performance14 Oct 2026 · www.linkup.so · Web pageLinkup states that it evaluated Tavily on SimpleQA using the same script used to measure its own search API. It reports a rounded Tavily score of 73%.

Company context

linkup.so

Linkup builds web search and retrieval infrastructure for AI developers and enterprise teams. Its API connects AI agents and language models to current web information for research, legal work, sales enrichment and other applications. Its main products are Fetch, which extracts content from individual webpages; Search, which returns web results and source-cited answers; and Research, which handles more complex investigations. Enterprise options include dedicated indexes, access controls and deployment within customers’ own cloud infrastructure.

Documented financing · private company

Amount$10MSeed
· Linkup press release Read follow-up source

Recent updates

7 updates

  1. Partnership

    Linkup announces search partnership with Aleph Alpha

    Linkup announced that its web search API powers Aleph Alpha’s internal tools and customer-facing AI solutions. The integration runs on EU infrastructure with zero data retention and supports industrial, public-sector and defense customers.

    Linkup

  2. Partnership

    Linkup partners with Cloudflare for AI Gateway web search

    Cloudflare launched its Web Search API with Linkup among its initial search partners. Developers can access live web results through AI Gateway with unified billing, observability and access controls.

    Cloudflare Blog

  3. Open model

    Linkup releases SPARSEUP retrieval model

    Linkup Research published an open sparse embedding model and its benchmark discussion.

    Linkup research post

More in Infrastructure

AWSBasetenBoardwalk