Octen adds $20 unlimited web search for developers, with queues at busy times
The subscription includes a limited image allowance. Pulse’s separate speed claim measures server processing, while its downloadable model weights restrict commercial use.
Loading page…
The subscription includes a limited image allowance. Pulse’s separate speed claim measures server processing, while its downloadable model weights restrict commercial use.
Listen to this story
Octen’s November 13 offer gives individual developers a $20 monthly subscription for unlimited Web Search, aiming to make frequent lookups more predictable to budget. It is not a guarantee of instant or all-inclusive access: Standard requests can queue, and full-content retrieval and other APIs may cost extra. The launch also promotes Pulse, but its API was already generally available on October 28, so the new development is the subscription—not a newly released search model.
Standard includes 1,000 image searches monthly; the $100 Plus plan includes 5,000 and priority over Standard requests.
Pulse’s 10-millisecond lite and roughly 20-millisecond standard figures measure server-side processing, not end-to-end response time.
Pulse weights and page-ID tables are CC BY-NC 4.0; commercial self-hosting requires a separate license, while inference code uses Apache 2.0.
Octen introduced a $20 monthly search subscription for individual developers on November 13, 2026, offering unlimited Web Search alongside its pitch for faster AI lookups. The announcement also promotes Pulse, a page-finding model the company says processes searches in milliseconds. But the subscription’s unlimited allowance does not cover every API, and requests can queue.
The subscription is the new commercial offer; Pulse itself has an earlier history. Octen’s October 28 technical announcement already described the Pulse API as generally available. That chronology separates the November pricing announcement from the underlying technology, whose speed, accuracy and licensing details were public more than two weeks earlier.
Pulse uses generative retrieval: a trained model produces a page identifier rather than repeatedly sorting candidates from a conventional search index. Octen says its four-billion-parameter version covers about 201 million pages. The company presents it as a way to reach destinations such as a software documentation page, pricing page or service-status website.
The headline 10-millisecond figure applies to lite mode, which returns a URL and title. Standard mode adds metadata and can supply relevant passages or page content; Octen puts its median processing time at about 20 milliseconds. Both figures measure server-side work, not the full wait for a network request, fetched content and an AI-generated answer.
Finding an existing address is also different from choosing the right one. In Octen’s own evaluation, Pulse placed the correct page first on 59.6% of navigational queries, versus 46.5% for its vector-retrieval baseline, which matches numerical representations of queries and pages. Octen acknowledges that rarely visited pages and pages published after training are better handled by a live index.
The November subscription brings a different proposition: predictable spending on repeated web searches. According to pricing details examined by VentureBeat, the $20 Standard plan includes unlimited Web Search and 1,000 image searches per month. It is a personal-account plan, with requests subject to queues during busy periods rather than guaranteed immediate processing.
Self-hosting brings another boundary. Although the November release calls Pulse open source, Octen’s technical announcement specifies CC BY-NC 4.0 for the downloadable weights and page-ID tables, limiting them to noncommercial use. The inference code uses Apache 2.0. Businesses seeking commercial deployment of the downloadable model are directed to contact Octen for licensing; downloading it is not blanket permission to run a commercial service.
Independent task testing offers a broader comparison, though it evaluates Octen Search rather than Pulse. Artificial Analysis’s Search Index, using November 9 data, keeps the answering model and test setup constant while changing the search provider. Octen’s highlights configuration scored 77, below the 80 achieved by Exa auto and Perplexity medium.
Octen’s task time was 15.2 seconds, compared with 19.0 for Exa lite and 39.3 for Parallel advanced. Those timings combine measured search time with derived model time. Its search and model costs totaled $23.57 per 1,000 tasks, against $42.15 for Exa lite and $58.60 for Parallel advanced. These are completed-task comparisons, not interchangeable prices for individual search calls.
For developers moving beyond personal use, Octen sells guaranteed capacity separately. Its Builder plan lists $2,099 monthly for up to 50 queries per second and a service-level agreement, with API usage billed separately. The low subscription price and the production commitment are different purchases.
Loading discussion...
Join the conversation
Explain when predictable spending would outweigh an unpredictable wait.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.