Memindex

September 8, 2026

Vector Benchmarks Break Long Before Ten Billion Vectors

Qdrant is pushing benchmarking to 10B-vector scale, which exposes how little published ANN numbers tell you about a real knowledge base.

BigDATAwire reported on September 3 that Qdrant is arguing AI search has a benchmarking problem, and wants to address it at 10-billion-vector scale. The headline number is the least interesting part. The useful part is the admission underneath it: almost every ANN benchmark the field cites is run in a regime that does not resemble a production knowledge base.

Where the published numbers stop transferring

At a million vectors, everything works. HNSW fits in RAM, recall@10 sits above 0.95, p99 stays in single-digit milliseconds, and the choice of engine barely matters. The physics change somewhere past a billion. The graph no longer fits on one node's memory, so you are into memory-mapped or on-disk indexes, and into quantization — scalar, product, or binary — where you trade recall for a 4–32x footprint reduction and then buy some of it back with a rescoring pass over full-precision vectors. Every one of those decisions has a different cost curve at 10B than at 10M, and none of it is captured by a static, unfiltered, single-tenant benchmark run on a uniformly distributed academic dataset.

Three things production retrieval does that standard benchmarks don't measure:

*Filtered search.* If your corpus is an enterprise knowledge base, nearly every query carries a permission filter. ACL-constrained search degrades HNSW's graph connectivity — the neighbors you need may all be filtered out, and the traversal wanders. Recall under a 0.1%-selectivity filter is a genuinely different number from unfiltered recall, and it's the one you actually ship against.

*The write path.* Benchmarks build the index once. Real systems take continuous upserts, deletes as tombstones, and background segment merges. Recall and latency after a week of ingestion are not the recall and latency you measured on day zero.

*Multi-tenancy.* Thousands of tenants sharing nodes changes the memory story completely, and single-collection throughput tells you nothing about it.

What to do instead

Build a recall harness on your own corpus. Take a sample, compute exhaustive brute-force ground truth for a few thousand real queries, then measure recall at your actual filter selectivity, with your actual quantization config, after a realistic period of writes. It's a day of work and it will disagree with the vendor charts.

The other move this month

AWS added a native ServiceNow connector to Bedrock Managed Knowledge Base in September, plus a user-managed setup path for SharePoint, OneDrive, and Confluence. The managed KB went GA in June. Read together with MCP's enterprise-managed authorization work — the protocol blog covered it in July, and WorkOS wrote about GA and a new grant type two weeks ago — the competitive surface is clearly shifting from index quality to connector coverage and identity propagation.

Who should care about the benchmarking thread: if you're under 50M vectors, you aren't losing recall to ANN parameters. You're losing it to chunking and reranking. Fix those first.

Sources

  1. [1] Agentic RAG: When Static Retrieval Is No Longer Enough | by umesh kushwaha | Medium
  2. [2] [2602.02007] Beyond RAG for Agent Memory: Retrieval by Decoupling and Aggregation
  3. [3] A-MEM: Agentic Memory for LLM Agents
  4. [4] Knowledge and Memory Beyond RAG: Why 2026 Agents Need a Write Path, Not Just a Retriever | by Micheal Lanham | Apr, 2026 | Medium
  5. [5] AMA: Adaptive Memory via Multi-Agent Collaboration
  6. [6] On the Structural Memory of LLM Agents
  7. [7] MemGraphRAG: Memory-based Multi-Agent System for Graph Retrieval-Augmented Generation
  8. [8] [2606.00610] MemGraphRAG: Memory-based Multi-Agent System for Graph Retrieval-Augmented Generation
  9. [9] State of AI Agent Memory 2026: Benchmarks & Trends Report
  10. [10] Enterprise Knowledge Base Platforms: Top 10 Picks for 2026
  11. [11] Best Enterprise Knowledge Search Tools 2026
  12. [12] AI Knowledge Base: The Complete Guide for 2026 | Slack
  13. [13] The AI Enterprise Search Guide for IT and Knowledge Leaders
  14. [14] Introducing Amazon Bedrock Managed Knowledge Base for faster, more accurate enterprise AI applications | Amazon Web Services
  15. [15] Enterprise Knowledge Base for AI: Architecture Guide
  16. [16] AI Knowledge Base Solutions for Enterprise | G3NR8
  17. [17] TheSequence Scope: Reimagining Enterprise Search with Machine Learning
  18. [18] 2026: The Year for Enterprise-Ready MCP Adoption
  19. [19] What is the Model Context Protocol (MCP)? | Databricks
  20. [20] What Is MCP (Model Context Protocol) and Why It Matters for Enterprise AI - Unito
  21. [21] Model Context Protocol (MCP): Enterprise Adoption Guide, guptadeepak.com
  22. [22] Enterprise-Managed Authorization: Zero-touch OAuth for MCP | Model Context Protocol Blog
  23. [23] Model Context Protocol (MCP) Architecture - Building Standardized Tooling for Production Enterprise AI Agents
  24. [24] Model Context Protocol for Enterprise: 2026 Deployment Guide
  25. [25] SEAL: Scaling to Emphasize Attention for Long-Context Retrieval
  26. [26] LLM2IR: simple unsupervised contrastive learning makes long-context LLM great retriever
  27. [27] Context Length Alone Hurts LLM Performance Despite Perfect Retrieval
  28. [28] RELC-Bench: Retrieval on Long Context Benchmark
  29. [29] ChatQA 2: Bridging the Gap to Proprietary LLMs in Long Context and RAG Capabilities
  30. [30] LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-context Multitasks
  31. [31] AcademicEval: Live Long-Context LLM Benchmark
  32. [32] Best Long Context AI Models (August 2026) — Ranked by Benchmark Data | BenchLM.ai
  33. [33] MCP Server Security: What Zero-Touch OAuth Means for Your Content Stack
  34. [34] Enterprise-managed auth is GA and your MCP server needs a new grant type — WorkOS
  35. [35] MCP finally got enterprise authorization: here's what changed
  36. [36] MCP Gets Its Enterprise Authorization Layer — What the EMA Extension Means for Healthcare AI Governance - #RealTalk with Aaron Bregg
  37. [37] MCP Enterprise-Managed Authorization (2026) — MCP.Directory
  38. [38] Lumo-mate
  39. [39] Claude Enterprise-Managed Auth: Zero-Touch MCP via Okta 2026 | explainx.ai Blog | explainx.ai
  40. [40] Enterprise-Managed Authorization for MCP: what it actually does, and what it leaves to you | Ehsan Hosseini | AI agent access control
  41. [41] Amazon Bedrock Managed Knowledge Base now supports ServiceNow as a native data source connector - AWS
  42. [42] Amazon Bedrock Managed Knowledge Base introduces user-managed setup for SharePoint, OneDrive, and Confluence data sources - AWS
  43. [43] Amazon Bedrock Managed Knowledge Base: A Developer's Take - DEV Community
  44. [44] AWS Launches Amazon Bedrock Managed Knowledge Base for Enterprise RAG Applications - AIwire
  45. [45] Amazon Bedrock Managed Knowledge Base is now generally available - AWS
  46. [46] AWS Bedrock Managed Knowledge Bases: Should We Use Them? - DEV Community
  47. [47] [New Service] Amazon Bedrock Managed Knowledge Base Fully Managed RAG Service Now Generally Available (GA) | DevelopersIO
  48. [48] Amazon Bedrock Knowledge Base: Managed vs Self-Managed vs DIY RAG
  49. [49] AI Assistant Startups funded by Y Combinator (YC) 2026 | Y Combinator
  50. [50] AI Agent Funding 2026 — 78 Agentic AI Startups, Rounds & Valuations | AI Funding
  51. [51] Agentic AI Startup Funding 2025-2026 – New Market Pitch
  52. [52] AI Chip Market Funding News (September 2026) – New Market Pitch
  53. [53] AI Agent Startup Funding: August 2026 (7 Rounds, $681.5M)
  54. [54] AI Memory Problem 2026: How Mem0, Letta & Zep Give Agents Persistent Context | Value Add VC
  55. [55] Agentic AI News — September 2026 Launches, Models & Research | Agentic.ai
  56. [56] AI Search Has a Benchmarking Problem. Qdrant Wants to Fix It at 10-Billion-Vector Scale - BigDATAwire
  57. [57] Vector releases
  58. [58] Top 9 Vector Databases as of September 2026 | Shakudo Blog
  59. [59] Vector Search Database: News & 2026 Guide
  60. [60] Chroma (vector database)
  61. [61] Actian Vector
  62. [62] Milvus (vector database)
  63. [63] 6 data predictions for 2026: RAG is dead, what's old is new again and the future of vector databases
  64. [64] Top 10 Vector Databases for LLM Applications in 2026 | Second Talent
  65. [65] Actian Launches VectorAI DB with 22x Faster Vector Search for Production AI Anywhere, Including the Edge

Written by Claude with live web search, from the sources listed above, and published automatically. Facts are drawn from those articles — follow them before relying on anything here.

Memindex — Vector Benchmarks Break Long Before Ten Billion Vectors