MongoDB Atlas Flex and Serverless Tiers
Parent: MongoDB Atlas · researched 2026-05-28T16:56:33.163Z· 12 sources · 11 concepts · skill mongodb-atlas-flex-serverless
MongoDB Atlas Flex tier (GA: February 6, 2025) is the successor to both the shared tier (M2/M5) and the serverless tier. As of January 22, 2026, M2/M5 clusters and Serverless instances are end-of-life
Overview
- MongoDB Atlas Flex tier (GA: February 6, 2025) is the successor to both the shared tier (M2/M5) and the serverless tier. As of January 22, 2026, M2/M5 clusters and Serverless instances are end-of-life and no longer supported - all existing instances were automatically migrated. Flex is now the entry-level paid cluster in Atlas. [source]
Flex Pricing Model
- Ops-per-second tiered hourly model capped monthly - no per-document billing: [source]
- Base: 100 ops/sec + 5 GB storage → $8/month [source]
- 200 ops/sec → $15/month [source]
- 300 ops/sec → $21/month [source]
- 400 ops/sec → $26/month [source]
- 500 ops/sec → $30/month (hard cap - no runaway billing) [source]
- Billed hourly, prorated. Burst to 500 ops/sec for 1 hour does not cost $30. [source]
- Key differences from Serverless: Fixed-price cap (no unbounded billing), no cold starts, always warm. [source]
Unsupported Features on Flex
- Private Endpoints (AWS PrivateLink, Azure Private Link, GCP PSC) [source]
- VPC/VNet Peering [source]
- Continuous Backup / PITR (daily snapshot only) [source]
- Database Auditing [source]
- Customer Key Management (BYOK) [source]
- Performance Advisor, Real-Time Performance Panel, Auto-indexing [source]
- allowDiskUse for aggregations [source]
- $where, mapReduce (no server-side JavaScript) [source]
- Sharded clusters (replica sets only) [source]
Supported Features on Flex (Added vs. old M2/M5)
- Atlas Search (full-text Lucene) [source]
- Atlas Vector Search (HNSW - with resource contention caveat) [source]
- Change Streams [source]
- Atlas Triggers and App Services [source]
- Full driver compatibility [source]
- Vector Search caveat: mongod and mongot share the same node on Flex. Resource contention causes higher query latency. Upgrade to M10+ with dedicated Search Nodes before production. [source]
Flex vs. Dedicated Decision Matrix
- Stay on Flex if ALL true: [source]
- Peak throughput < 500 ops/sec [source]
- Data < 5 GB [source]
- Connections < 500 [source]
- No private networking requirement [source]
- No PITR backup requirement [source]
- No BYOK encryption requirement [source]
- Development, staging, prototype, or low-traffic production [source]
- Move to Dedicated (M10+) when ANY trigger: [source]
- Approaching 500 ops/sec consistently [source]
- Data exceeds 5 GB [source]
- Need private endpoints or VPC peering [source]
- Need PITR (compliance requirement) [source]
- Need BYOK encryption at rest [source]
- Atlas Vector Search going to production [source]
- SLA requires predictable low-latency [source]
- Need > 500 concurrent connections [source]
- Need custom MongoDB version pinning [source]
- Cost break-even: Flex max = $30/mo; M10 = ~$57/mo. For light workloads (<100 ops/sec, <1 GB): Flex saves 85%+ vs. M10. [source]
Terraform
- Note: Do NOT use MongoDB::Atlas::FlexCluster in CloudFormation - use MongoDB::Atlas::Cluster instead. [source]
Kubernetes Operator
- Upgrade to AKO 2.12.0+. Use spec.flexSpec in AtlasDeployment CRD. To upgrade Flex → Dedicated: spec.upgradeToDedicated: true. [source]
Migration Path: Flex → Dedicated
Serverless History (Deprecated)
- Serverless used RPU (Read Processing Unit) / WPU (Write Processing Unit) per-document billing. Caused runaway charges for unindexed queries. Cold starts after ~5 minutes idle (30-60 second reconnect delay). Replaced by Flex due to unpredictable billing, cold start latency, and feature gaps. [source]
Anti-Patterns
- Flex for production with PrivateLink requirement: Must use M10+ for private endpoints [source]
- Production Vector Search on Flex: Resource contention causes latency spikes [source]
- Expecting PITR on Flex: Daily snapshots only [source]
- Pinning MongoDB version on Flex: Flex auto-upgrades - use dedicated for version pinning [source]
- Overbuilding M10 for dev/staging: Flex handles dev/staging at 1/7th the cost [source]
Free Tier (M0) vs. Flex
- M0 auto-pauses after 30 days idle (cold start on resume). Flex never pauses - more suitable for always-on dev/staging environments. [source]
References
Children
- Atlas Flex Pricing Model (frontier)
- Atlas Flex Technical Limits (frontier)
- Serverless RPU WPU Billing (frontier)
- Serverless Cold Start Behavior (frontier)
- Serverless Deprecation and EOL (frontier)
- M2 M5 Migration to Flex (frontier)
- Flex vs Dedicated Decision Matrix (frontier)
- Flex Tooling Migration Terraform K8s CLI (frontier)
- M0 Free Tier vs Flex Comparison (frontier)
- Atlas Flex Unsupported Features (frontier)
- Flex Cost Break-Even Analysis (frontier)
Frontier under this node: Atlas Flex Pricing Model, Atlas Flex Technical Limits, Atlas Flex Unsupported Features, Flex Cost Break-Even Analysis, Flex Tooling Migration Terraform K8s CLI, Flex vs Dedicated Decision Matrix, M0 Free Tier vs Flex Comparison, M2 M5 Migration to Flex, Serverless Cold Start Behavior, Serverless Deprecation and EOL, Serverless RPU WPU Billing