OctenAI

Today @OctenAI is releasing Octen Pulse, web search at the speed and cost of inference. The model is the index, so one search takes ~10 ms, about one LLM token. We're open-sourcing it with a tech report so you can test it yourself. Link below.

A new Harvard and Stanford paper evaluated 10 frontier LLMs against 26 embedding models across a 37-task benchmark. Octen-8B ranked #1 among all embedding models at 77.2, a statistical tie with the top LLM: Gemini 3.1 Pro, at 77.6. The cost gap is striking: $154 vs. $0.11 per benchmark pass (1,431x). The paper's conclusion is a division of labour, not a replacement: embedding models for classification, similarity, and clustering; LLMs for reasoning-intensive retrieval. We think that's the right read. You can read the whole analysis here: arxiv.org/pdf/2608.12875

Our Web Search API is the cheapest. On top of that, we are now giving you the full text of the top 10 URLs by default, completely for free, on every search.  We believe having the full context is a fundamental need for AI agents, so it should just be the default. If you build agents, your typical calls ask for 10 results or fewer, so with this feature you may just stop paying for page content entirely. Beyond the first 10, the price is only $0.50 per 1,000 results.

We just benchmarked Octen Extract against 11 other fetch APIs across 100 pages. The result: #1 overall with an 88% success rate. We are not the single best on every individual aspect, and the report shows that openly. But we took the top spot by holding up everywhere rather than just winning one category. Why is that important? Your AI reads every kind of website, so your provider's blind spots instantly become your own. That is why consistent performance across all page types is better than perfection in just one. And the most unique features of Extract: Every result includes page structure and category metadata, giving your agent useful context about the page without requiring it to process the full content first.  And you only pay for successful fetches. The full evaluation report and reproduction scripts are live here: bki.sh/WUeFPsP

@OctenAI's Model Gateway is live on Product Hunt today! One API key for every frontier model. Add one line and your favorite frontier model searches the live web itself with no tool definition, no handler & no loop. I will be in the comments answering everything. Get 15% off your rebates and tell us how we're doing.

Introducing Octen Model Gateway. One API key. Models and search together, with the fastest responses in the world. Thousands of developers have joined in the last 2 weeks. To say thanks, everyone gets a 15% rebate on all consumption. Try it now in the comments: