Octen
Category: AI Search / Retrieval
Octen builds a real-time, LLM-native web search API and retrieval models designed for AI agents, with low latency and very high concurrency. Octen was founded in 2025. The company is led by Kuan Zou. Based in San Francisco, United States. Total funding raised: $10M. Latest round: Seed. Key investors include Square Peg (lead), Argor, AI scientist angels.
- Founded
- 2025
- Headquarters
- San Francisco, United States
- Total funding
- $10M
Value proposition
Search infrastructure built for machine consumption rather than human browsing: agents can fire thousands of concurrent queries and get structured, fresh results in milliseconds.
Products and solutions
Web Search API, Image Search, Video Search (early access), News/Business/Place Search, Extract (URL to markdown), Octen-Embedding models (0.6B/4B/8B, open-sourced on HuggingFace), VL Embedding (multimodal), Answer, Search Agent, Deep Research, Grounded Generation. Delivered via API, SDKs
Unique value
Claims a ~10ms page-location model, P50 latency of 62ms holding flat from 1K to 1M+ QPS, and content indexed within 5 minutes of publication. Also offers #1-ranked retrieval embeddings (RTEB, MMEB-V2).
Target customer
AI agent builders, frontier AI labs, LLM application developers, and hardware companies; individual developers via the $20/month unlimited plan.
Industries served
AI agents and LLM applications, developer tools, research assistants, chatbots, enterprise AI
Technology advantage
Proprietary distributed search engine with a trillion-scale, multilingual index and real-time indexing. In-house embedding models that topped the RTEB leaderboard (Octen-Embedding-8B, 0.7953) and the MMEB-V2 leaderboard (octen-vl-embedding-large, 80.09). Claims #1 for speed and #2 for cost in the Artificial Analysis Search API Benchmark. SOC 2 Type 2.
How they differentiate
Positions itself as the all-rounder on speed, cost and quality. In the Artificial Analysis benchmark Octen highlights scored 77 at $0.024 per task and 13.6s per task, against Exa at $0.064 and 34.7s and Brave LLM context at $0.079 and 20.0s. Competes on high concurrency, freshness and flat-rate unlimited pricing.
Main competitors
Exa, Tavily, Brave Search API
Key partnerships
Live integration with Claude (MCP/Skills), available in Claude Code, Cursor, VS Code and other MCP clients. The company says it works with frontier AI labs, agent builders and hardware companies, and with design partners across the generative AI ecosystem (none named).
Major milestones
2025 founded, Jan 2026 Octen-Embedding-8B tops the RTEB leaderboard, Mar 2026 web search API launch and $10M seed led by Square Peg, live on Claude (Mar 2026), May 2026 octen-vl-embedding-large tops MMEB-V2, Nov 2026 $20/month unlimited plan and multimodal image search launch.
Growth metrics
Claims 1M+ QPS on a single account, 62ms P50 latency and sub-5-minute indexing. Embedding model topped the RTEB leaderboard within weeks of development starting. The API was in invitation-only beta at the April 2026 launch.
Market positioning
Search layer for the agentic internet; low-latency, high-concurrency infrastructure for AI agents.
Geographic focus
Global; dual hubs in San Francisco and Singapore
Patents and IP
No patents found. Open-source Octen embedding models on HuggingFace; Octen-Team skills on GitHub.
About Kuan Zou
Ex-Alibaba Cloud, led AI search for 5+ years, serving hundreds of millions of end-users; earlier built Baidu's enterprise search platform from scratch; Nanyang Technological University (Singapore).
Latest news about Octen
More AI Search / Retrieval companies
Official website: https://octen.ai