macrostack
Migration guide · LLM Gateways & Routing

The 4 best OpenRouter alternatives

OpenRouter is a hosted router for language models: one API key and one OpenAI-compatible endpoint in front of 600+ models from 80+ providers, with routing, fallbacks and one bill. There is nothing to deploy; you buy credits and spend them at the providers' own token prices.

90
Bottom line

A good front door to many models, priced as a fee on your spend rather than a subscription. Once the fee is larger than the cost of running a gateway, LiteLLM or Bifrost does the routing on your own machine with your own keys.

Jump to the full comparison →

The cost

No subscription. The Free plan opens 25+ free models at 50 requests a day (1,000 a day once you have bought 10 credits). Pay as you go adds no markup on inference and charges a 5.5% platform fee on credit purchases ($0.80 minimum; 5% paid in crypto). Business is 8%, with EU or US in-region routing; Enterprise is by contract, with fee discounts. Bringing your own provider keys carries no fee on the first $25,000 of list-price inference a month, then 5%. Read on openrouter.ai/pricing and its FAQ on 17 Nov 2026.

Is OpenRouter free? The exact limits

Why people consider an alternative

Three reasons. The fee: 5.5% of every credit purchase is a charge on all your model spend, so $10,000 a month of tokens carries $550 of fees, more than a small server running a gateway of your own. The data path: every prompt and completion passes through a third party, which a regulated or client-confidential workload may not allow. And the dependency: a router in front of every call is one more service whose outage stops your product while the model providers are up.

When OpenRouter is still the right call

Stay if you want to try many models this afternoon without opening an account at each provider, if your spend is small enough that the fee costs less than the time to run a gateway, or if its free models are enough for a prototype. Routing and fallback across providers come configured, where a self-hosted gateway makes you set them up, and the Business plan adds in-region routing for EU or US data rules.

AlternativeLicenseSelf-hostPricingSovereignty
LiteLLM★MIT (core); enterprise/ directory under a separate commercial licenceYesFree and self-hosted; you pay the model providers directly. Enterprise features such as SSO and audit logs are sold separately.90
Direct SDKs (no framework)Your code (official SDKs: MIT / Apache-2.0)YesFree; you pay each model provider directly.93
BifrostApache-2.0YesFree and open source (Apache-2.0); you pay the providers directly. Enterprise edition by contact, with a 14-day trial.88
Vercel AI GatewayProprietary (hosted service)NoNo markup and no platform fee on tokens; payment-processing fees are passed on. Free tier with a subset of models and rate limits. Add-ons such as regional hosting (5%) cost extra.30
90
Macrostack's top pick

LiteLLM

The open gateway most teams run: 100+ providers, your keys, no fee on spend.

Which one fits your situation?

There is no single best answer — it depends on what you are actually trying to do.

Just tell me the best OpenRouter alternative

LiteLLM90

Our top pick — the highest Sovereignty Score at 90, and the one we would choose ourselves.

LiteLLM vs Direct SDKs (no framework) →

I have no budget — is there a genuinely free one?

Direct SDKs (no framework)93

Open source under Your code (official SDKs: MIT / Apache-2.0). Free; you pay each model provider directly.

Direct SDKs (no framework) vs LiteLLM →

I want to self-host it and own the data

Bifrost88

Self-hostable, so your data stays on your own server. Free and open source (Apache-2.0); you pay the providers directly. Enterprise edition by contact, with a 14-day trial.

Bifrost vs LiteLLM →

Every alternative, compared

#1★ TOP PICK

LiteLLM

The open gateway most teams run: 100+ providers, your keys, no fee on spend.

90
OPEN SOURCEMIT (core); enterprise/ directory under a separate commercial licenceSELF-HOSTLOCAL-FIRST

A proxy with an OpenAI-compatible endpoint in front of 100+ providers, with virtual keys, per-team budgets and fallbacks. You run it and pay the providers directly, so nothing is added to your token spend. GitHub showed 63,375 stars and a push on 17 Nov 2026. The core is MIT; the enterprise/ directory is under a separate commercial licence.

Strengths

  • +No fee on token spend: providers bill you at their own price
  • +Prompts never leave your infrastructure
  • +Virtual keys, per-team budgets and fallbacks built in

Trade-offs

  • −You run it: deployment, upgrades and uptime are yours
  • −SSO and audit logs sit behind the commercial licence
Free and self-hosted; you pay the model providers directly. Enterprise features such as SSO and audit logs are sold separately.
#2

Direct SDKs (no framework)

Call each provider's own API: no router in the path, no fee on top.

93
OPEN SOURCEYour code (official SDKs: MIT / Apache-2.0)SELF-HOSTLOCAL-FIRST

With two or three models, the simplest alternative is no router at all: each provider's own SDK and API key, billed at that provider's price with nothing added. You give up one bill and automatic fallback across providers, and take one service out of every request.

Strengths

  • +Nothing between your code and the provider
  • +Each provider's price, with no fee on top

Trade-offs

  • −One account, key and bill per provider
  • −Fallback between providers is code you write
Free; you pay each model provider directly.
#3

Bifrost

An Apache-2.0 gateway in Go, started with one command.

88
OPEN SOURCEApache-2.0SELF-HOSTLOCAL-FIRST

An open-source gateway from Maxim AI: one OpenAI-compatible API across 23+ providers and 10,000+ models, with fallbacks, load balancing, semantic caching, budgets, virtual keys and MCP support. It starts with npx -y @maximhq/bifrost or a Docker image. GitHub showed 9,645 stars and a push on 17 Nov 2026. Maxim calls it completely free and open source, with an enterprise edition by contact.

Strengths

  • +Apache-2.0 throughout, with no separate commercial directory
  • +Semantic caching, budgets and virtual keys built in
  • +One command to start, or a Docker image

Trade-offs

  • −Younger and smaller than LiteLLM: 9,645 stars against 63,375
  • −You run it, and the enterprise edition is a sales conversation
Free and open source (Apache-2.0); you pay the providers directly. Enterprise edition by contact, with a 14-day trial.
#4

Vercel AI Gateway

A hosted gateway with no markup on tokens.

30
SOURCE-AVAILABLEProprietary (hosted service)

Vercel's hosted gateway: one endpoint and one bill across many providers, with no markup and no platform fee on tokens by its pricing docs (updated 8 Sep 2026). The free tier covers a subset of models with rate limits; on paid use you can bring your own keys at zero markup, and you pay the payment-processing fees. Some add-ons cost extra, among them a team-wide model allowlist, regional hosting at 5%, and bring-your-own-key without fallback.

Strengths

  • +No markup on tokens
  • +Nothing to run
  • +Bring your own keys at zero markup on paid use

Trade-offs

  • −Hosted and proprietary: every prompt passes through Vercel
  • −Add-ons such as regional hosting carry a surcharge
No markup and no platform fee on tokens; payment-processing fees are passed on. Free tier with a subset of models and rate limits. Add-ons such as regional hosting (5%) cost extra.

Questions people ask

▸ Is OpenRouter cheaper than calling the providers directly?

No, but the gap is small and visible: it passes the providers' token prices through without markup and charges 5.5% when you buy credits (8% on Business). On $1,000 of model spend that is $55. Calling providers directly costs the token price alone, in exchange for an account and a bill at each one.

▸ What is the best self-hosted alternative to OpenRouter?

LiteLLM, for most teams: the most widely used open gateway, 100+ providers behind an OpenAI-compatible endpoint, with budgets and virtual keys, and no fee on spend because you pay the providers directly. Bifrost (Apache-2.0, written in Go) is the newer option of the same shape, with semantic caching built in. Both mean running the gateway yourself.

▸ Does Vercel's AI Gateway charge a fee?

Not on tokens, by its pricing docs: no markup and no platform fee, with payment-processing fees passed on. Some add-ons cost extra, such as regional hosting at 5%. It is a hosted service like OpenRouter, so prompts still pass through a third party.

▸ Why are Portkey and Helicone not on this list?

Both changed hands or direction this year. On 17 Nov 2026 portkey.ai read 'Portkey is now PRISMA AIRS AI Gateway', and the last commit to its open-source gateway was on 25 May 2026; helicone.ai read 'Helicone is now part of Experiential Labs. Details to come soon.' They stay on the Portkey alternatives page; they are left off this one until their direction is clear.

Compare them head-to-head

Related comparisons

Entry last verified 2026-11-17. Licenses and pricing change — spotted something out of date? That's a correction we want.

The Macrostack brief

New swaps, worth your inbox.

A short, occasional email when we add a high-intent alternative or ship a new head-to-head. No spam, no selling your address — unsubscribe in one click.