Octen builds web search for AI systems rather than for people. The startup has now released a model called Octen Pulse and says it returns text results in 10 milliseconds, well under the time it says most search services for AI take.
Octen describes Pulse as open source but has not said which licence applies. The company announced the launch from San Francisco on 13 November 2026.
Alongside the launch, Octen cut the price of its paid plan. Individual developers can now run unlimited searches on its service for $20 a month.
Search Built For Machines
AI agents are software that carries out tasks on its own, and they often need to look things up on the web while they work. They do this through an API (application programming interface), a link that lets one program send requests to another and get structured answers back.
Octen argues that speed matters more for agents than for people, because an agent may run many searches in a single task. The company says most search services for AI take more than 300 milliseconds to answer each request.
Big tech firms are also giving agents more of their own tools. In October, Google unveiled Gemini agents that get their own company email addresses.
Pulse uses a method called generative retrieval. Octen says that instead of checking a stored list of web pages, as a traditional search engine does, the model has absorbed millions of web addresses from large sites and can go from a question to a matching page in one pass.
Octen says Pulse can run on the device itself, which it argues suits robots, smart glasses and smartphones that need answers without waiting on a distant server.
The company’s documentation adds detail. It gives 10 milliseconds as the median server response in a “lite” mode and 23 milliseconds in standard mode, with up to 10 web addresses per query. It also says Pulse is best for well-known sites such as official pages and help centres, and points users to its regular web search for news and fresh pages.
Octen’s Other Search Products
Pulse sits alongside Octen’s main search service, Octen Fast. The company says half of Fast’s answers arrive within 62 milliseconds, the median time, on SealQA Hard. Octen claims that makes Fast more than four times quicker than its nearest rival, which it did not name.
SealQA is a test set built by researchers at Virginia Tech in 2025. It uses fact-seeking questions where web searches turn up conflicting, noisy or unhelpful results, and its hardest set has 254 questions.
Fast can also handle more than one million queries a second, according to the company. A second service, Octen Premier, covers more specific searches, for example for businesses, news and places. Octen has also added image search to its paid plan.
What Independent Tests Show
Octen points to results from Artificial Analysis, an independent firm that tests AI products. Its search benchmark, a standard test that scores products side by side, runs the same AI model and agent software against each search service and changes only the search provider.
On the firm’s current leaderboard, with data to 9 November 2026, Octen’s search service is the fastest of 26 products from 14 providers. It takes 0.21 seconds per query and 15.7 seconds per task. Octen’s own announcement gave 15.1 seconds per task.
Speed is not the same as quality. On the leaderboard’s overall quality score, Octen scores 77 and ranks seventh of the 12 providers listed. Artificial Analysis tested Octen’s search service, not the new Pulse model.
Octen also says its search cost $9.07 per 1,000 tasks, against $30 to $66 for other providers in September tests. That figure covers the search fee only, and the AI model adds a further $14.50. The latest data shows at least one provider, DeepSeek, with a lower search fee, at $8.47.
Artificial Analysis also notes that speed per call is only part of the picture. A fast service can still slow a task down overall if the AI model needs to search it more often to reach an answer.
Who Is Behind Octen
Octen raised $10M earlier this year in a seed round, a startup’s first sizeable outside funding. Australian venture firm Square Peg led the round, and Singapore-based Argor and a group of AI researchers also invested.
The company was about three months old when it announced the round in April 2026. It has offices in Singapore and San Francisco and says it has built up more than 5,000 active developers since opening its service to all users in July.
Founder and chief executive Kuan Zou previously led AI search at Alibaba Cloud and built enterprise search at Baidu. Octen says its staff come from companies including Meta, Google, TikTok, Alibaba, Baidu and DeepSeek. In January, the company says its embedding model, which turns text into numbers so computers can compare meaning, topped RTEB, a public leaderboard for finding relevant documents.
Voices On Speed And Cost
Zou links the bet on speed to earlier shifts on the internet. “Every time the internet got faster, a wave of new applications followed, and search for AI works the same way,” he said. He argued that faster retrieval lets companies bring AI products to market sooner.
Cost matters too for teams building agents. “For agents that rely on the web, search can sometimes cost more than the model itself,” said Sid Sharma, head of go-to-market, the team that handles sales and launches, at AI company Inception.
Aixin Sun, associate dean of the College of Computing and Data Science at Nanyang Technological University in Singapore, compared a large language model, the kind of AI system behind chatbots, to “a powerful brain”. “Intelligence becomes far more powerful when information is abundant, current, and immediately available,” Sun said.
Octen is pitching its new model and lower price at developers building agents and AI devices that need quick web answers. Independent tests back its speed claims for its main search service, but not yet for Pulse, and they rank Octen seventh of 12 on answer quality.
