Octen launched Octen Pulse, an open source model that it says returns search results in 10 milliseconds, so AI models can search while they think.
Key Takeaways
- Octen Pulse is a generative retrieval model that returns search results in 10 milliseconds
- A new $20 a month subscription gives individual developers unlimited API calls
- The subscription now also includes image search for agents
- Octen says it has grown to 5,000+ active developers since announcing general availability in July
The company announced that Pulse is open source and joins Octen Fast and Octen Premier as the third search method in its lineup. Octen describes itself as the search engine built for machines.
Also Read: AI Agents That Keep Their Own Wiki Remember More Than Rivals, Hong Kong And Tencent Researchers Find
Why 10 Milliseconds Matters
Octen says that at this speed, search stops being a separate step a model pauses for and becomes part of how it reasons. Each Pulse search takes about the same time and cost as generating one token.
The company says most AI search APIs take more than 300 milliseconds per call. A model finishes a reasoning step, stops, searches and waits. Octen says Pulse is 30x faster than most AI search APIs.
Pulse has learned millions of URLs across major websites, so a single inference turns a question directly into a matching result. Because it is open source and can run on devices, Octen says it also brings real time search to robots, smart glasses and smartphones that can't wait on a round trip to the cloud.
Three Ways to Search
Octen Fast is the core search API. The company says it returns structured results at a median of 62 milliseconds on the SealQA Hard benchmark, more than 4x faster than its nearest rival. It supports more than 1 million queries per second.
In the Artificial Analysis Search API Benchmark, Octen was the fastest provider at 0.21 seconds per search query and 15.1 seconds per task, according to the company.
| Octen Method | What It Does |
|---|---|
| Pulse | Open source generative retrieval model with results in 10 milliseconds |
| Fast | Core search API with a median of 62 milliseconds on SealQA Hard |
| Premier | Business search, news search and place search through the API and console |
Unlimited Search for $20 a Month
Octen says most AI search APIs charge per call, and agents search dozens of times per task, so the bill climbs with every step. Developers then let agents search less, and the answers get worse. For $20 a month, individual developers get unlimited API calls.
The company cites Artificial Analysis, which it says ran the same model and harness against 26 search API products from 13 providers. In its September results, Octen's search cost $9.07 per 1,000 tasks, compared with $30 to $66 for other providers.
"We're building search the way utilities deliver water and electricity, so developers never have to ration it as they grow," said founder and CEO Kuan Zou.
Sid Sharma, head of go-to-market at Inception, said Octen "makes search spend easier to predict."
Image Search and the Team
The subscription now includes image search. Octen says it gives image generation models real visual references and lets agents build answers that combine text and images.
Zou previously led AI search at Alibaba Cloud and enterprise search at Baidu. His team comes from Meta, Google, TikTok, Alibaba, Baidu, DeepSeek and Xiaohongshu. Octen is headquartered in San Francisco and is backed by Square Peg and Argor. Developers can get an API key at octen.ai.
Read Next: OpenAI’s Latest V7 Gives Agents a Memory Layer on GPT-5.6
Originally published on Fathom News.
