AI & ML
Best OpenRouter Alternatives (2026): 7 LLM Gateways Compared
Seven OpenRouter alternatives compared: LiteLLM, Bifrost, Portkey, Helicone, Vercel, Cloudflare and Requesty, with fees, licenses, BYOK and free tiers.
Your tool hereOn tool pages, blog posts and the feeds, next to what people came to read.$19 / 3 months →If you want to stop paying OpenRouter's fee on credits, the cheapest swap is a gateway you host yourself: LiteLLM, Bifrost or Portkey's open-source gateway, all free to run with your own provider keys. If you want one hosted bill with no markup on tokens, Vercel AI Gateway and Helicone's hosted gateway both say they add 0%. Requesty is the closest like-for-like hosted swap, with a flat 5% markup.
Every fee, license and limit below was checked against each vendor's own page or GitHub repository on 11 November 2026.
What OpenRouter does, and what it charges
OpenRouter puts 600+ models from many providers behind one API on its paid plans. Its FAQ calls it "a drop-in replacement for OpenAI", and when a provider returns an error it falls back to the next provider automatically.
The money side, from the pricing page:
- No markup on inference. Models are billed at each provider's list price.
- A fee on credit purchases: 5.5% on the Standard plan, 8% on Business. Enterprise fees are set by contract.
- BYOK (your own provider keys): the first $25,000 a month of list-price inference has no fee; above that, OpenRouter charges 5% of what the request would have cost on OpenRouter.
- Free plan: 25+ free models, 50 requests a day and 20 a minute, rising to 1,000 a day once you have bought at least $10 of credits. No BYOK on Free.
So the cost of OpenRouter is the credit fee, not the token price. People look elsewhere when that fee adds up, when they need their own logging and budgets, or when traffic has to stay on their own servers. If you only want OpenRouter for its free models, see free AI API tiers instead; a gateway does not make models free.
1. LiteLLM — the open-source proxy most people start with
Best for: teams that want to run their own gateway and pay providers directly.
LiteLLM is a Python SDK plus a self-hosted proxy
server that lets you call 100+ LLM APIs in the OpenAI format. The repository is MIT
licensed, except an enterprise/ directory under its own commercial license. The proxy
handles virtual keys, spend tracking, guardrails and load balancing, and its
reliability docs cover retries and
fallbacks. Response caching works with
Redis/Valkey, in-memory, disk, S3, GCS and semantic backends.
The proxy docs describe it as self-hosted (Docker, Helm or Terraform); SSO, audit logs and multi-team management are in the enterprise tier, priced through sales. To try it locally:
uv tool install 'litellm[proxy]'
litellm --model gpt-4o2. Bifrost — a self-hosted gateway focused on speed
Best for: self-hosters who want a web UI and semantic caching out of the box.
Bifrost is an Apache-2.0 gateway from Maxim. Its
README lists 23+ providers behind one OpenAI-compatible API, with automatic failover, load
balancing across keys and providers, and semantic caching. The
docs add budgets and rate limits at virtual-key,
team and customer level. Clustering, guardrails and in-VPC deployments are enterprise
features; no price is published. Start it with npx -y @maximhq/bifrost or
docker run -p 8080:8080 maximhq/bifrost.
3. Portkey — open-source gateway, hosted logs
Best for: teams that want an MIT gateway now and hosted observability later.
Portkey's gateway is MIT licensed and runs
locally with npx @portkey-ai/gateway. Its README covers 1,600+ models, retries and
fallbacks you can trigger on specific errors, and simple and semantic caching; the examples
pass your own provider key. The hosted product is billed by recorded logs, not tokens:
pricing gives a free Developer plan with 10k logs a month
(3-day retention) and Production at $49/month for 100k logs. Going over the log limit does
not block requests; the extra logs are just not recorded.
One thing to know: portkey.ai now says Portkey "is now PRISMA AIRS AI Gateway", aimed at enterprises. Check the roadmap before you build on the hosted side.
4. Helicone AI Gateway — observability first, 0% markup
Best for: teams that want request logs, sessions and cost tracking with the gateway.
Helicone started as an Apache-2.0
LLM observability platform (more in
LLM observability and evaluation tools with a free tier)
and now runs a hosted
gateway at ai-gateway.helicone.ai. Its gateway docs
say you can "use credits with 0% markup" or bring your own keys, with automatic failover,
caching and an OpenAI SDK base-URL swap. The pricing page
has a free Hobby plan (10,000 requests, 7-day retention) and Pro at $79/month. A separate
Rust ai-gateway repository exists for
self-hosting, but it is marked public beta and has had no commits since November 2025.
5. Vercel AI Gateway — one hosted bill, zero markup
Best for: teams that want a hosted gateway with no fee on tokens, on Vercel or not.
Vercel AI Gateway is a hosted API at
ai-gateway.vercel.sh that accepts OpenAI Chat Completions, OpenAI Responses and Anthropic
Messages requests. Your app does not have to run on Vercel. Its
pricing page says it "charges no markup and no
platform fee on tokens", including with BYOK, though you pay any payment processing fees.
BYOK needs purchased credits. You get provider and model fallbacks, request logs, budgets
per team, project or key, and prompt caching. Some add-ons cost extra (custom reporting, a
team-wide provider allowlist or zero data retention).
The free tier gives a monthly credit on a subset of models, with lower rate limits; our free AI API tiers roundup has the amount and the card requirement.
6. Cloudflare AI Gateway — free core features, on Cloudflare's edge
Best for: apps already on Cloudflare that want caching and analytics in front of their existing provider keys.
Cloudflare AI Gateway adds analytics, logging, caching, rate limiting, retries and model fallbacks in front of providers such as OpenAI, Anthropic, Gemini and Workers AI. The pricing page says the core features are free. If you buy credits through Unified Billing, a 5% fee applies to the purchase; token prices pass through without markup. Gateways created before 24 September 2026 keep 100,000 logs in total on Workers Free, 10,000,000 per gateway on Workers Paid; newer ones use Workers Logs pricing.
You can store provider keys in Cloudflare (BYOK) so your app stops sending them. There is an OpenAI-compatible endpoint, though Cloudflare now marks it deprecated for single-model chat calls and points to its REST API instead.
7. Requesty — the closest hosted swap
Best for: people who like OpenRouter's model and want a simpler fee structure.
Requesty puts 600+ models behind one OpenAI-compatible endpoint. Pricing is a flat 5% on top of what you spend, with no platform or seat fee. The free plan (no card) gives the free models, 200 requests a day, routing, caching, fallbacks and spend tracking. Pay-as-you-go adds BYOK, spend limits and fuller observability; the page does not list a separate BYOK fee.
Side by side
| Gateway | Hosting | License | Fee on top of tokens | Free tier | OpenAI API | BYOK |
|---|---|---|---|---|---|---|
| OpenRouter | Hosted | Closed | 5.5% on credits (8% Business) | 25+ free models, 50 req/day | Yes | Yes, 5% above $25K/month |
| LiteLLM | Self-host | MIT (enterprise dir excluded) | None | Free to run | Yes | Yes (your keys) |
| Bifrost | Self-host | Apache-2.0 | None | Free to run | Yes | Yes (your keys) |
| Portkey | Self-host or hosted | MIT (gateway) | None on tokens; hosted billed by logs | 10k logs/month | Yes | Yes (your keys) |
| Helicone | Hosted (self-host in beta) | Apache-2.0 (platform) | 0% on credits | 10,000 requests | Yes | Yes |
| Vercel AI Gateway | Hosted | Closed | None | Monthly credit, subset of models | Yes | Yes, needs purchased credits |
| Cloudflare AI Gateway | Hosted | Closed | 5% on Unified Billing credits | Core features free | Yes (deprecated for single-model) | Yes (stored keys) |
| Requesty | Hosted | Closed | 5% on spend | Free models, 200 req/day | Yes | Yes (pay-as-you-go) |
"Free to run" means no license fee; you still pay for the server and the providers.
How to choose
- You want zero fees and can run a container: LiteLLM, the most-starred here (63K GitHub stars, 100+ providers); Bifrost for a built-in web UI, semantic caching and zero-config setup.
- You want one hosted bill with no markup: Vercel AI Gateway. Helicone also says 0% and suits you if logs and sessions matter more than the gateway itself.
- You already have provider keys and just want caching and analytics: Cloudflare AI Gateway, free for the core features.
- You want OpenRouter, but cheaper per dollar: Requesty's 5% is below OpenRouter's 5.5% Standard fee. The gap is small; compare the models you actually use.
- You might self-host later: Portkey's MIT gateway lets you start locally and keep the same API if you move to the hosted logs.
- You use mostly BYOK under $25,000 a month: OpenRouter itself charges no BYOK fee up to that level, so switching may save less than you think.
FAQ
Is there a free, open-source alternative to OpenRouter?
Yes. LiteLLM (MIT, with an enterprise directory excluded), Bifrost (Apache-2.0) and Portkey's gateway (MIT) are all free to self-host. You bring your own provider keys and pay the providers directly, so there is no fee on top of tokens.
Does OpenRouter mark up model prices?
No. OpenRouter bills inference at each provider's list price and charges a fee when you buy credits: 5.5% on Standard and 8% on Business. BYOK requests are free up to $25,000 a month of list-price inference, then 5%.
OpenRouter vs LiteLLM?
OpenRouter is a hosted service with one bill and its own credits. LiteLLM is software you run yourself: no fee, but you need your own provider accounts and you operate the server, database and updates.
Which LLM gateways have an OpenAI-compatible API?
All eight in this post. Cloudflare's OpenAI-compatible endpoint still works but is marked deprecated for single-model chat calls, in favour of its REST API.
Gateways route requests; they do not make models free. For providers with a recurring free allowance, see free AI API tiers that actually last, and the current list on the free credits page. For more AI tooling, browse the AI & ML directory.