LiteLLM Review: Pricing & Comparison
The de facto open-source AI gateway: free to self-host, unifying 140+ providers and 1,800+ models behind one OpenAI-compatible API
Last verified: 2026-10-15 · Visit official site →
It is not a relay — it is what you build a relay with
Start with the category. LiteLLM (litellm.ai) is an open-source AI gateway (a Python SDK plus a proxy server) with a paid enterprise edition. Its own gateway page says “Self-hosted, on-prem, or in your cloud”, tagged On-prem / Multi-Cloud / Kubernetes / Helm — there is no managed SaaS version; you deploy it on your own machine or cloud.
That means it solves a completely different problem from a relay. A relay answers “I do not have an official account or channel, or the official channel is impractical from where I am”. LiteLLM answers “I already have many official keys — how do I unify them behind one API with rate limiting, retries, cost accounting and access control?” If your problem is the first one, LiteLLM cannot help; if it is the second, it is effectively the default answer in this category.
Scale: 1,800+ models, 61k stars
The homepage claims one OpenAI-compatible API to 140+ providers and 1,800+ models; the gateway page says 1,892. The documentation names Bedrock, Azure OpenAI, OpenAI, Anthropic, Vertex AI, vLLM and NVIDIA NIM directly.
One of those numbers can be checked independently: the GitHub repository BerriAI/litellm has 61k+ stars and 1,000+ contributors, and it had commits on the day we verified. It also maintains a public model pricing database (models.litellm.ai) that can be used directly for cost calculations — genuinely useful for budgeting.
Worth noting: the site’s own scale figures are inconsistent (100+ vs 140+ providers, and 53K / 58.8K / 61.2K stars appearing in different places), which is not especially rigorous for a product built around governance.
Pricing: free open source, capacity-priced enterprise
- Open source: $0, labelled Free forever · self-hosted
- Enterprise: no published price. The site's own wording is that pricing is sized to annual gateway
request capacity, deployment architecture and support needs, and is never per token —
quote only, via sales or AWS Marketplace / resellers
- Enterprise support SLAs: Sev 0 = 1 hour, Sev 1 = 6 hours, Sev 2-3 = 24 hours
(7am-7pm PT, Mon-Sat), vulnerabilities = 72 hours
- Trial: a 30-day enterprise trial key is available by emailThe company also claims SOC 2 Type 2 and ISO 27001 (trust.litellm.ai). Performance claims — such as ~4.5x more requests per dollar, or the cost reductions cited in customer stories — are vendor or customer claims and were not independently verified by us.
Who it’s for
It suits teams already using several official APIs who need a unified gateway layer (key management, virtual keys and budgets, rate limiting and retries, logs and cost dashboards), and it works well as the foundation of a self-built relay — several of the relays catalogued on this site are in fact built on One API / New API forks, and LiteLLM is the most widely used open-source implementation of the same idea.
It does not suit users who have no upstream keys and simply want to buy credit and start calling, nor individual developers who do not want to run a server. For those, the managed options below are the better fit.
Information verified 2026-10-15. Pricing, model counts and SLA windows reflect what litellm.ai, its enterprise page and the GitHub repository showed on that date.
Related reviews
- Cloudflare AI Gateway: free AI traffic control layer, caching/rate-limiting/monitoring, edge acceleration
- Portkey: the full LLMOps suite, 1600+ model coverage, 50+ guardrail rules, a production-grade AI gateway
- OpenRouter: a cross-vendor multi-model aggregation platform, 200-300+ models, one API key to call OpenAI/Anthropic/Google/Meta/Mistral and more
- Vercel AI Gateway: 0% markup, $5 free credit, zero data retention (ZDR), a go-to for frontend developers
Quick facts
| Pricing model | The open-source edition is free to self-host (the site labels it Free forever · self-hosted at $0); Enterprise pricing is not published — the site states it is sized to annual gateway request capacity, deployment architecture and support needs, and is never per token — quoted via sales or through AWS Marketplace; published support SLAs are Sev 0 in 1 hour, Sev 1 in 6 hours, Sev 2-3 in 24 hours (7am-7pm PT, Mon-Sat) and 72 hours for vulnerabilities |
|---|---|
| Model coverage | The site claims one OpenAI-compatible API to 140+ providers and 1,800+ models (the gateway page says 1,892 models); the docs name Bedrock, Azure OpenAI, OpenAI, Anthropic, Vertex AI, vLLM and NVIDIA NIM, and a public model pricing database is maintained at models.litellm.ai |
| Latency / SLA | The open-source edition carries no SLA; the enterprise edition publishes response windows by severity (Sev 0 in 1 hour, Sev 1 in 6 hours) and offers a 30-day enterprise trial key; performance claims such as ~4.5x more requests per dollar and customer cost-saving figures are vendor or customer claims |
| Mainland direct connect | Self-hosted, so there are no vendor-side nodes; upstream reachability depends on your own network |
| Best for | Developers / Enterprise |
| Referral program | No public affiliate or rebate programme found. |
Pros
- Genuinely open source and genuinely free: a self-hosted gateway adds no markup, and your keys and traffic never pass through a third party (the FAQ states plainly that they do not see your data or traffic)
- Project activity is independently verifiable — 61k+ GitHub stars, 1,000+ contributors, commits on the day we checked — and the model pricing database is public
- The enterprise edition publishes response-time SLAs by severity (Sev 0 in one hour), and day-zero support for new models is among the fastest in this category
Cons
- It is infrastructure, not a relay: you need your own upstream API keys and a server, so it does nothing to solve having no official account or channel
- Enterprise pricing is fully opaque and sales-gated, which makes budgeting unpredictable
- The site contradicts itself on scale (100+ vs 140+ providers, 53K vs 58.8K vs 61.2K stars), which is sloppy for a governance-focused product
Compare more AI API relays
See the full comparison board — filter by price tier, model coverage, and mainland direct-connect status.
Back to the comparison board →