# Atlas Pricing Programs, Cost Governance and TCO > Source-anchored research for Atlas Pricing Programs, Cost Governance and TCO. Run status: COMPLETED. Parent: [MongoDB Atlas Cost Optimization](https://llms-explorer.com/tree/mongodb-atlas-cost-optimization/) · 7 facets · 43 facts · page: https://llms-explorer.com/tree/atlas-pricing-programs-cost-governance-and-tco/ ## Definitions - Existing non-compliant clusters are not fixed or blocked. They can only be changed in ways that bring them into compliance, and are flagged via the /orgs/{ORG-ID}/nonCompliantResources endpoint. Existing autoscaling limits outside a new policy can still scale automatically, but manual scaling beyond policy limits is blocked. (high confidence) — [source](https://www.mongodb.com/en-us/docs/atlas/atlas-resource-policies/) - Decision heuristic repeated across sources: Atlas is the default for small teams with limited DBA/ops staff (MongoDB staff on the community forum call it the most cost-effective option for DBA/ops headcount); self-hosting fits large deployments with dedicated DBAs, compliance/data-residency or air-gapped needs, or scale where Atlas cost is prohibitive. (medium confidence) — [source](https://oneuptime.com/blog/post/2026-03-31-mongodb-atlas-vs-self-hosted-cost-management/markdown) ## Structure and components - Neither the MongoDB announcement nor press coverage publishes Agent Engine unit rates, billing units or free-tier terms. Agent Memory uses Voyage AI embeddings, so token-metered embedding cost is likely part of it. Memory and governance can be adopted without the runtime. (medium confidence) (verified as of 2026-10-02) — [source](https://www.mongodb.com/de-de/company/newsroom/press-releases/mongodb-launches-atlas-agent-engine-to-put-ai-agents-in-production-without-a-new-stack) - Atlas Cost Explorer filters and groups usage by organization, project, cluster and service. It shows six months of usage and up to 18 months of billing data, and spans linked organizations under cross-organizational billing. This is the native Inform-phase visibility layer. (medium confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/docs/atlas/architecture/current/billing-data/) ## How it works - Cost governance pattern: model Flex spend as a monthly bound ($8-$30 per cluster) so dev/test fleets have a known worst case. A fleet of N always-busy Flex clusters costs at most $30N, so consolidating onto one shared M10 (about $57) only wins when N is 2 or more clusters that each sit near the cap and fit the Dedicated feature need. This is derived from list prices, not a vendor statement. (low confidence) — [source](https://mongodb.com/docs/atlas/billing/atlas-flex-costs) - Example cost cap: forbid cluster.modify when min tier is below M30 or max tier is above M60; forbid diskSizeGB above 4096; forbid clustersInProject above 2. This blocks provisioning and manual or auto scaling outside the range. (high confidence) — [source](https://www.mongodb.com/en-us/docs/atlas/atlas-resource-policies/) - The Framework lifecycle is iterative: Operate outputs inform changes to the Scope, Inform updates the Scope, Optimize sets refinements as targets, then the cycle repeats. The 2025 update dropped 'cloud' from domain and capability names, for example 'Optimize Usage and Cost'. (low confidence) — [source](https://finops.org/insights/2025-finops-framework) - Atlas resolves tags at invoice export time, not when the cost occurs. A cluster deleted before export shows as untagged, which breaks tag-based allocation for ephemeral clusters. A customer request filed 2026-06-24 (status Started) got a MongoDB reply on 2026-07-15 saying it is being evaluated for prioritization. Workaround: export invoices before deleting short-lived clusters, or keep an external tag-to-cluster mapping. (low confidence) (verified as of 2026-07-15) — [source](https://feedback.mongodb.com/ideas/CUSTOMER-I-10624) - Atlas-only features (Atlas Search, Vector Search, Data Federation, Triggers, Charts, performance advisor, online archive) have no self-hosted equivalent in Community; equivalent search needs a separate cluster such as Elasticsearch. Search, triggers and device sync use Atlas-specific APIs, which raises migration-off cost. (medium confidence) — [source](https://oneuptime.com/blog/post/2026-03-31-mongodb-atlas-vs-self-hosted-cost-management/markdown) ## Measurements and reference values - Flex pricing is a banded ops/sec model: $8 base (0-100 ops/sec, $0.0110/hr), then marginal $7 (to $15, $0.0205/hr), $6 (to $21, $0.0288/hr), $5 (to $26, $0.0356/hr), $4 (to $30, $0.0411/hr) for 400-500 ops/sec. Billed hourly, prorated. (high confidence) (verified as of 2026-10-18) — [source](https://mongodb.com/docs/atlas/billing/atlas-flex-costs) - Flex monthly cost is capped at $30. The base price includes 5 GB storage (uncompressed BSON plus indexes), 100 ops/sec and unlimited data transfer. Atlas does not charge Flex clusters for outgoing data transfer. (high confidence) (verified as of 2026-10-18) — [source](https://mongodb.com/docs/atlas/billing/atlas-flex-costs) - MongoDB's worked example: 100 ops/sec for 20 days, 250 ops/sec for 5 days, 500 ops/sec for 5 days costs about $13.67 (about $5.28 if deleted after the first 20 days). Because billing is hourly and prorated, bursts cost only for the hours spent in the higher band. (medium confidence) — [source](https://www.mongodb.com/blog/post/dynamic-workloads-predictable-costs-mongodb-atlas-flex-tier) - Flex replaced Shared and Serverless: new M2/M5 and Serverless creation ended Feb 2025; as of Jan 22, 2026 existing M2/M5 and Serverless instances were migrated (M2/M5 to Flex; Serverless under 5 GB to Free or Flex, larger or with serverless private endpoints to Dedicated). Migrated clusters now bill on ops/sec bands. (high confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/docs/atlas/create-database-deployment.md) - Atlas cost guidance pairs detective controls with IaC levers: consistent tags (team, env, cost center) feed Cost Explorer; dev/test clusters can be paused (paused=true) or terminated via Terraform or the Admin API on a schedule; autoscale-down happens at most once per six hours; manual scale-down is suggested when usage stays below about 45% over 30 days. (medium confidence) — [source](https://mongodb.com/docs/atlas/architecture/current/cost-saving-config) - Prosopo reported cutting an Atlas bill of over $3,000/month (about $1,000 M40 instance, $700 continuous cloud backup, $1,000 internet data transfer) to about $160/month on one dedicated Hetzner server for roughly 500GB. The stated 90% matches neither figure exactly; $3,000 to $160 is about 95%. Staffing time was not reported. (low confidence) (verified as of 2025-11-12) — [source](https://prosopo.io/blog/we-cut-our-mongodb-costs-by-90-percent/) ## Problems, failure modes and limitations - Triggers to leave Flex for Dedicated (M10+): storage beyond 5 GB; sustained throughput at or above 500 ops/sec (pricing tiers stop there and a hard limit applies, with a sub-second per-connection cooldown that can queue operations over a second); need for dedicated hardware, consistent resources, or granular backup configuration. (high confidence) — [source](https://mongodb.com/docs/atlas/billing/atlas-flex-costs) - Resource policies bound configuration, not total spend. No Atlas control found hard-stops billing. Policies only limit cluster shape, so spend can still grow through number of allowed clusters, storage use within the cap, or data transfer. (low confidence) — [source](https://www.mongodb.com/en-us/docs/atlas/atlas-resource-policies/) ## Comparisons and alternatives - The cheapest Dedicated tier, M10, lists at $0.08/hr (pricing page headline: starts at $56.94/month), versus the Flex ceiling of $30/month. At list price Flex is cheaper at every load it can serve, so there is no cost break-even; the switch is driven by limits and features. The cost step is roughly 1.9-2x with no intermediate tier. (high confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/pricing) - Illustrative comparison: a 3-node Atlas M30 replica set is about $1,180/month before storage and transfer, versus about $440/month compute for three AWS r6g.xlarge self-hosted nodes. Adding assumed ops labor of 4-8 h/month at $150/h ($600-1,200) gives roughly $1,040-1,640/month self-hosted, so the gap closes or inverts for small teams. Backup, monitoring and patching are listed but not priced, and the labor figures are assumptions, not measurements. (medium confidence) (verified as of 2026-03-31) — [source](https://oneuptime.com/blog/post/2026-03-31-mongodb-atlas-vs-self-hosted-cost-management/markdown) - The Prosopo savings came with reduced resilience and scope: a single standalone server (not a replica set), daily cron mongodump backups to a storage box, self-run VPN/firewall/monitoring, and a workload that is ML batch rather than real-time. A like-for-like replica set with PITR would cost more, so the 90% figure is not a general benchmark. (medium confidence) — [source](https://prosopo.io/blog/we-cut-our-mongodb-costs-by-90-percent/) - Atlas bill drivers beyond instance size are backup storage (PITR extra), internet/cross-region egress (ingress is free) and add-ons such as Search; in the Prosopo case backup plus data transfer was about 55% of the bill. Cost modelling should itemize these rather than compare VM prices. (medium confidence) — [source](https://www.percona.com/blog/mongodb-pricing-explained-what-youre-really-paying-for/) ## Facts and statements - Atlas Embedding and Reranking API is pay-as-you-go: text models bill per token, multimodal models per pixel. Billing is set at organization level, and charges go to the Atlas account linked to the model API key. (high confidence) — [source](https://www.mongodb.com/docs/voyageai/management/billing/) - Current text embedding list prices per 1M tokens: voyage-4-large $0.12, voyage-4 $0.06, voyage-4-lite $0.02, voyage-code-4 $0.12. voyage-finance-2, voyage-law-2 and voyage-code-2 are $0.12. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/voyageai/management/billing/) - Free tier is 200M tokens for most current models and 50M for voyage-finance-2, voyage-law-2 and voyage-code-2. Older models (voyage-3-large and voyage-code-3 at $0.18, voyage-3.5 at $0.06, voyage-3.5-lite at $0.02) get no free tokens. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/voyageai/management/billing/) - Automated Embedding in Atlas Vector Search bills the same per-token model rates. Tokens are metered at index creation (initial sync), on document inserts and updates, and on every $vectorSearch query. The free allowance is 200M tokens per model, one-time, shared across all projects and clusters in the org, and it does not refresh. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/vector-search/crud-embeddings/automated-embedding/billing/) - Cost visibility: Cost Explorer can filter by AI model, and Automated Embedding has its own Service filter value. Metering-derived costs refresh every 24 hours. Invoices break out usage per model and per project. (medium confidence) — [source](https://www.mongodb.com/docs/voyageai/management/billing/) - Rate limits (RPM and TPM) are set at org level and can be lowered per project. Usage tier rises with cumulative spend: add a payment method for Tier 1, $100 for Tier 2, $1,000 for Tier 3. A linked child org takes the higher of its own and its parent's tier. Automated Embedding rate limits apply per cluster instead. (medium confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/voyageai/management/billing/) - Atlas Agent Engine was announced 2026-09-29 and is in public preview. Atlas Agent Runtime and Atlas Agent Memory are consumption-priced, and usage counts against existing Atlas commitments, so no new contract is needed. (high confidence) (verified as of 2026-10-02) — [source](https://www.mongodb.com/de-de/company/newsroom/press-releases/mongodb-launches-atlas-agent-engine-to-put-ai-agents-in-production-without-a-new-stack) - Multimodal Voyage models (voyage-multimodal-3.5 and 3) cost $0.12 per 1M text tokens plus $0.60 per billion pixels, with 200M text tokens and 150B pixels free. (low confidence) (verified as of 2026-10-19) — [source](https://docs.voyageai.com/pricing) - Rerank prices are $0.05 per 1M tokens for the standard reranker and $0.02 for the lite one. The Voyage pricing page names the models rerank-3 and rerank-3-lite, while search summaries cited rerank-2.5. The model name needs checking before it goes in a TCO model. (low confidence) (verified as of 2026-10-19) — [source](https://docs.voyageai.com/pricing) - Flex does not support multi-region/multi-cloud, sharding, global clusters, network peering, private endpoints, LDAP/auditing, or point-in-time restore; only select cloud regions are available. These gaps force a Dedicated move regardless of cost. (medium confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/atlas/create-database-deployment.md) - Atlas Resource Policies are org-level, Cedar-based, forbid-only rules (permit is not supported). With no policies, users can do anything their role allows. (high confidence) — [source](https://www.mongodb.com/en-us/docs/atlas/atlas-resource-policies/) - Cost-relevant policy dimensions include min/max cluster tier and autoscaling bounds (minGeneralClassInstanceSizeValue, maxGeneralClassInstanceSizeValue), disk size (diskSizeGB), clusters per project (context.project.clustersInProject), topology, shard count, cloud provider and region. The tier cap does not cover Low CPU or NVMe SSD classes. (high confidence) — [source](https://www.mongodb.com/en-us/docs/atlas/atlas-resource-policies/) - Terraform resource mongodbatlas_resource_policy takes org_id, name, optional description and a policies list whose body holds Cedar text. It is imported as {ORG_ID}-{POLICY_ID}. Policies can be committed and run through CI. Atlas docs also list the Admin API, UI and CloudFormation. Only Organization Owners can create, update or delete policies. (high confidence) — [source](https://www.pulumi.com/registry/packages/mongodbatlas/api-docs/resourcepolicy) - Gotchas: duplicate policy names return 400 Bad Request; one policy per .cedar file; a minimum shard count above 1 needs a !context.cluster.isConvertingToSharded exception or replica-set to sharded conversions are blocked; TLS policies need Terraform provider 1.33.0 or later. (medium confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/en-us/docs/atlas/atlas-resource-policies/) - Project-level billing alert: Alerts > Add Project Alert > Billing > 'Monthly pending invoice ($) total is...' with an operator, threshold and notification channels. It tracks cumulative billed amount for the current calendar month. It notifies only and does not stop spend. The Atlas cost docs recommend an alert at a set amount (example $100). (medium confidence) (verified as of 2026-01-30) — [source](https://feedback.mongodb.com/ideas/CUSTOMER-I-10398) - FinOps Framework 2025 was approved in March 2025. It adds Scopes, defined as a segment of technology-related spending to which practitioners apply FinOps concepts. The initial Scopes are Public Cloud, SaaS and Data Center. Scopes are not mutually exclusive. (low confidence) — [source](https://finops.org/insights/2025-finops-framework) - Atlas resource tags (for example department, project or cost center) appear on invoice line items. The Return All Pending Invoices endpoint exposes them in results.lineItems.tags. MongoDB recommends feeding that data into external FinOps tools for chargeback, by environment, team or tag value. (high confidence) — [source](https://www.mongodb.com/docs/atlas/architecture/current/billing-data/) - Atlas provides billing data in FOCUS 1.3 format. The FinOps Open Cost and Usage Specification gives a vendor-neutral schema. Reports come from the Invoices page (CSV FOCUS Cost & Usage Report) or from the FOCUS Data API's asynchronous request and retrieve endpoints. Columns include BilledCost, EffectiveCost in billing currency, and SubAccountName in place of groupName. (medium confidence) (verified as of 2026-10-18) — [source](https://www.mongodb.com/docs/atlas/billing/focus-api/) - Third-party FinOps platforms ingest Atlas billing directly. DoiT's MongoDB Intelligence module, announced 2025-11-07, breaks spend down by organization, project and SKU, with trend tracking and an AI assistant. (low confidence) (verified as of 2025-11-07) — [source](https://www.doit.com/changelog/mongodb-intelligence-finops-visibility-for-atlas-built-into-doit-cloud-intellige) - Atlas Well-Architected cost guidance maps to the Optimize phase: enable auto-scaling on the cluster tier to cut over-provisioning, and use tags plus programmatic billing pulls for allocation. MongoDB does not publish an explicit Inform/Optimize/Operate mapping, so this mapping is the researcher's inference. (low confidence) — [source](https://www.mongodb.com/docs/atlas/architecture/current/billing-data/) - Community Edition lacks built-in hot backups/PITR, LDAP/Kerberos, TDE and audit logging (Enterprise-only), has no official Kubernetes operator and no official support. Self-managed teams own replica set failover, monitoring, backup scripts, tuning and upgrades. Enterprise Advanced is a per-server/per-core subscription that adds support and Ops Manager, with licensing rising at scale; Percona (a competing vendor) offers open-source alternatives. (medium confidence) (verified as of 2025-04-14) — [source](https://www.percona.com/blog/mongodb-pricing-explained-what-youre-really-paying-for/) - Self-managing gives control Atlas withholds: Prosopo could not kill leaked connections from the Atlas admin shell and had to stop services, while on its own server it manages connections directly. (low confidence) — [source](https://prosopo.io/blog/we-cut-our-mongodb-costs-by-90-percent/)