Linkup✓

Introducing SPARSEUP: the first model release from @Linkup_platform's Research team, led by @thibault_formal. SPARSEUP is the missing sparse companion of DenseOn and LateOn. Same backbone, same data, <150M, 56+ on BEIR-13. The model is open-source, use it now on @huggingface. Blog: Model: (Apache 2.0)

RT @RespanAI: Everything is a plugin: providers, MCP, routing, fallbacks. Each one takes a few lines in config. Here’s an example with OpenAI and @Linkup_platform, a web search integration we natively support: [providers.openai] vendor = "openai" key = "env:OPENAI_API_KEY" [tools.linkup] vendor = "linkup" key = "env:LINKUP_API_KEY" Switch on the tool loop, and GPT can search the web with Linkup. Your app stays a plain chat call.

Join us in Paris for an evening on Retrieval. Speakers from Linkup Research, @cohere and @Sorbonne_Univ_ , dinner and drinks. 29/10, Linkup Paris HQ RSVP on Luma, link in comments.

@cohere @Sorbonne_Univ_

One of our early crawl runs hit a test site with no delay between requests. Within minutes it was returning 500s. We’d knocked it offline. Now Mercator, our crawler, visits each site at most once a second. To still read billions of pages a day, going fast meant going wide. Read the full story from our CTO Denis Charrier, link in comments.

How we built Mercator:

We're a tech partner for the @hcompany_ai x @42AI_ Hackathon in Paris, 7–8 November. Brief: build an agent that operates desktop software and completes a real task like a human (or better). Stack: Linkup provides the web search layer: real-time, granular data from authoritative sources, retrieved in one API call. H Company brings their agents and models, payments by @stripe, and @gladia_io for AI voice infra. Prizes: €10,000 in cash, plus credits. More rewards to be announced. Selection is competitive. Apply at link in comments.

@hcompany_ai @42AI_ Apply:

Every team evaluating Linkup asks the same first question: how do you perform against LLM native search and the competition? Fair question. On public benchmarks, we lead: SimpleQA and SealQA-0, state of the art. But raw accuracy only tells you so much. That's why we launched verticalized benchmarks that map to real use cases: GTM (People, Company), Legal (RedlineBench), Finance (FinSearchComp). Each links to an open repo with the dataset, queries, and methodology. Re-run it yourself. But what really matters is what works for you. Picking a provider on a public leaderboard alone is how you end up with failures in production: a benchmark measures general performance, not yours. So we go one step further and help you build a custom eval on your own queries and data, guide included. Public benchmarks show who leads in general. A custom eval shows who leads for you.

Access the benchmarks:

RT @philippemizrahi: Hiring a second 🇺🇸founding GTM for @Linkup_platform. We have a state-of-the-art product in a white hot market, and grew 10x in the last 6 months, so @stephksong and I need some help to manage the inbound. If you've sold technical products as the underdog we should talk.

RT @philippemizrahi: For the last couple of years we've been quietly building Mercator, the crawler behind @Linkup_platform's search API. It now pulls billions of pages a day. Finally wrote up how.

High-reliability inference needs high-performance retrieval.

@SambaNovaAI is building one of the leading inference platforms for enterprises, helping organizations deploy frontier open models with the performance, security, and operational reliability required for production AI.

We’re excited to make @Linkup_platform seamlessly available on the @SambaNovaAI platform, enabling developers to enrich those models with fresh, verifiable information from the web.

@hcompany_ai is building one of the most advanced computer-use agent platforms in Europe.

Their agents click, type, scroll, interact with interfaces and complete tasks autonomously. At production scale, that requires a search layer that's fast, reliable, and fully compliant. With Linkup's web search API, H Company powers live retrieval across its entire agent stack, giving every agent access to grounded, real-time web context exactly when it's needed.

Check out our full user story

The @Linkup_platform team is in London for one of the most exciting builder events of the summer: @unicorn_mafia’s Hardware Hackathon, alongside @AnthropicAI, @ElevenLabs, and @modal.

Join us:

Thanks to @seedcamp for hosting!