AI cost · LiteLLM
LiteLLM cost: what you actually pay
LiteLLM never charges per token itself — it's a free, open-source proxy that passes every call straight through to whichever provider (Bedrock, Azure, Vertex, OpenAI, Anthropic) you point it at; the only LiteLLM-branded cost is an optional, unlisted Enterprise license plus whatever compute you spend self-hosting it.
Public pricing
| Plan | Price | Billing | Includes |
|---|---|---|---|
| Open Source | FreeMIT-licensed proxy; unlimited self-hosted use at no license cost. | usage | 100+ provider integrations, virtual keys, load balancing, spend tracking/budgets, fallbacks, request/response logging; SSO free up to 5 users |
| Enterprise | CustomNot publicly listed; requires a sales conversation (30-day trial available). Third-party estimates cite roughly $250-$2,500/month depending on scale, unconfirmed on LiteLLM's own site. | custom | SSO/SCIM, OIDC/JWT auth, audit logs, key/team-based guardrails, secret manager integration, key rotation, 24/7 support |
LiteLLM has no token markup of its own — every $/MTok figure a team sees is whatever Bedrock/Azure/Vertex/OpenAI/Anthropic charges directly; LiteLLM sits in front as a routing, budgeting, and observability layer. Self-hosting it still costs real infrastructure (compute, a Postgres DB for spend logs/virtual keys, often Redis for caching/rate limiting) on top of the $0 license. SSO is free up to 5 users on the open-source tier; beyond that requires the paid Enterprise license.
Last updated: 2026-07-22. Sourced from LiteLLM Enterprise (official) ↗, LiteLLM Docs: Enterprise (official) ↗.
Where the real LiteLLM cost hides
The sticker is the smallest part of the story. For LiteLLM, these are the line items that quietly inflate the bill:
- LiteLLM is free, so teams underestimate the real cost of running it: compute + Postgres + Redis + ops time for a production-grade deployment is not $0 even though the license is.
- Because LiteLLM proxies dozens of providers transparently, teams lose track of which upstream (Bedrock vs. Azure vs. direct Anthropic) is actually cheapest for a given model — the gateway makes switching easy but doesn't force a rate comparison.
- Budgets, team/key rate limits, and guardrails that actually prevent overspend are stronger in the paid Enterprise tier — open-source-only deployments at scale are how teams discover a runaway API key after the invoice, not before.
- SSO beyond 5 users, audit logs, and key rotation require the Enterprise license — orgs needing SOC2-style controls end up paying for Enterprise for compliance features, not for higher throughput.
What an audit finds on LiteLLM
Teams self-host LiteLLM assuming it's 'free,' under-invest in budgets/team limits (manual on open-source, native on Enterprise), and the actual waste shows up entirely on the underlying Bedrock/Azure/Vertex/OpenAI bill that the gateway was supposed to help control.
LiteLLM pricing FAQ
Does LiteLLM charge per token or per API call?
No — LiteLLM is a proxy; you pay your model provider (Bedrock, Azure, Vertex, OpenAI, etc.) directly at their rates. Any LiteLLM-specific charge is a flat Enterprise license fee, not usage-based.
Is the open-source version really free for production use?
Yes — it's MIT-licensed and self-hosted with no usage caps; you pay only for your own infrastructure and the underlying model APIs.
What does LiteLLM Enterprise cost?
It's not publicly listed. LiteLLM requires contacting sales for a quote and offers a 30-day trial; third-party estimates range from a few hundred to a few thousand dollars per month depending on scale, unconfirmed by LiteLLM directly.
What is LiteLLM really costing you?
Spendassay connects your LiteLLM spend to actual usage and engineering output — a proof-level AI cost report with the wasted dollars named. Neutral across every AI vendor.
Free · read-only · no card
Other ai infrastructure pricing
More: the complete AI cost management guide · every tool, with prices · cost per outcome · compare platforms · how we measure