There's a pattern emerging in the AI infrastructure market, one that became harder to ignore when Octen AI stepped out of stealth mode recently with $10 million in seed funding. The San Francisco and Singapore-based startup is building what it calls AI-native search infrastructure—search engines designed not for people typing queries into browsers, but for large language models and autonomous agents pulling information on the fly.
Square Peg led the round, with Argor Capital Management joining in. Perhaps more telling than the funding itself: on January 12, Octen's embedding model had claimed the top spot on the RTEB leaderboard, a benchmark that evaluates how well models handle retrieval and embedding tasks. In the infrastructure layer beneath AI applications, that leaderboard has become something of a proxy for technical credibility.
The timing feels deliberate. Just days before announcing the funding, Octen launched its web search API with performance specifications that reveal where the company sees its opening. According to the company, median latency sits at 62 milliseconds for search queries—quick enough for real-time agent interactions. The system can reportedly handle over a million queries per second on a single account, with index updates arriving within five minutes of publication. (The company's own documentation cites a 99-millisecond average response latency, reflecting a different statistical slice of the same performance data.)
Under the hood, Octen's offering centers on three API endpoints: embedding, web search, and web chat. The Python SDK hit PyPI the same day the funding went public, suggesting the team had been preparing for a coordinated launch.
But the real story might be what's happening on that leaderboard. Octen's Octen-8B embedding model posted a mean task score of 0.8045 on RTEB in early January, a result the company detailed in a technical blog post discussing domain-specific training strategies and optimizations across both public and private benchmark tracks. The model family includes smaller 4B and 0.6B parameter variants alongside the flagship version—a product lineup that hints at flexibility for different deployment scenarios.
The founder brings relevant pedigree. Kuan (Colin) Zou spent more than five years leading AI Search at Alibaba Cloud before launching Octen, and before that built Baidu's enterprise search platform. LinkedIn lists the company's headcount somewhere between 11 and 50 employees, operating under the Singapore entity APITECH AI PTE. LTD.
What Octen is chasing—this notion that AI agents need fundamentally different search infrastructure—attracted serious capital last year. Exa, another search API targeting AI agents, raised $85 million at a $700 million valuation in September, with Benchmark leading. The same month, You.com closed a $100 million round at a $1.5 billion valuation, positioning itself as agent infrastructure while processing what it claimed was over a billion queries monthly.
The pattern suggests investor conviction, or at least investor hope, that a new category is forming. Traditional search engines optimized for human users—with their emphasis on ranking, snippets, and page experience—may not be what an autonomous agent needs when it's pulling facts mid-conversation or assembling information across dozens of sources in milliseconds.
Octen plans to deploy its seed capital across four areas: product development, developer adoption, enterprise partnerships, and team expansion on both the engineering and developer relations fronts. The company hasn't disclosed revenue figures, named customers, or the valuation at which it raised. Its status page shows uninterrupted uptime across all services since launch—a clean record, though admittedly a short one.
Whether the company can maintain that technical edge as competitors scale and larger players take notice remains an open question. For now, Octen is betting that being early—and being fast—gives it a foothold in a market still figuring out what it wants.
