How Does Parallel Work With Google Cloud?

Parag Agrawal says Parallel is working with Google Cloud as a search and grounding provider for enterprise agent APIs. On Google Cloud, web search attached to Gemini models or other models, to agents, to chat apps, or to other inference, is either Google search or Parallel search. Parallel search is product integrated.

He does not think a pre-training crawl is particularly useful for a fresh web index. Parallel crawls pages model companies skip for pre-training, because a page a model already trained on is not useful. Model companies are not patient enough to wait for slow random JavaScript, since the tokens per compute come back one order of magnitude, or two orders of magnitude, too poor. They take X trillion tokens instead. Parallel is completionist, so the extra effort is worth it. For nine out of 10 use cases, buying language models to get work done also means wanting the web and search infrastructure built for agents. Model companies can build the infrastructure, buy it, or partner.

Parallel wants model-company partnerships and will make them. No other partnership can be named yet. The Google Cloud role was announced the day of the conversation, after time with technical teams, training teams, product teams, and commercial teams so Gemini models used with Parallel return exceptional results.