The honest short list: LiteLLM if you want to self-host and pay nothing per token, Requesty if you want a hosted router with a simpler 5% markup and native Claude Code support, Portkey if you want guardrails and governance, Helicone if observability is the real requirement, Vercel or Cloudflare if you are already on that platform and want zero markup, direct provider keys if you call one model, and Continuum if the workload is a coding agent rather than an application.
- Zero-fee hosted options exist. Vercel AI Gateway charges no markup and no platform fee on tokens.
- LiteLLM is the self-host answer: no per-token fee, real operational cost instead.
- Requesty charges a flat 5% markup on inference with no top-up fee and no seats.
- Cloudflare AI Gateway is free for analytics, caching, and rate limiting; its unified billing adds 5%.
- If you call one model, the best alternative to a gateway is the vendor SDK.
- For coding agents, the comparison is a plan versus a meter, not a fee versus a fee.
Work out what you are actually replacing
OpenRouter does four distinct things: it aggregates a catalogue, it normalizes an API, it fails over between providers, and it bills you in one place. Almost nobody wants to replace all four. Start from the complaint.
| The complaint | Go look at |
|---|---|
| The 5.5% credit fee | Vercel AI Gateway (0%), direct keys, LiteLLM self-hosted |
| Another data processor in the chain | LiteLLM, Portkey OSS, Helicone self-hosted |
| The extra network hop | Direct provider keys, or a gateway inside your own VPC |
| No guardrails or PII controls | Portkey, Requesty enterprise, LiteLLM enterprise |
| Cannot see what any of it cost by team | Helicone, Portkey, Requesty, Continuum analytics |
| Only one model, all day | The vendor SDK. Stop reading. |
| A coding agent, not an app | Continuum, or a client-side router |
The eight options, with real numbers
| Option | Hosting | What it charges |
|---|---|---|
| OpenRouter | Hosted | 5.5% on credit purchase ($0.80 min); 0% markup on tokens |
| LiteLLM | Self-host | $0 OSS. Enterprise is quote-only. You pay for infra and ops. |
| Requesty | Hosted | 5% markup on inference. No subscription, no seats, no minimum. |
| Portkey | Both | OSS free unlimited. Cloud: free 10k logs, $49/mo for 100k, +$9/100k |
| Helicone | Both | Hobby free (10k requests). Pro $79/mo, Team $799/mo |
| Vercel AI Gateway | Hosted | No markup, no platform fee. Add-ons metered separately. |
| Cloudflare AI Gateway | Hosted | Core features free. Unified billing adds 5% on credits. |
| Direct provider keys | None | List price. Zero overhead, N integrations. |
LiteLLM: the self-hosted default
An open-source Python SDK plus a proxy that speaks OpenAI format in front of 100+ providers, with virtual keys, spend tracking, budgets, fallbacks, caching, guardrails, and an admin UI. 56,762 stars on GitHub as of 19 August 2026. It removes the per-token fee entirely and removes the third-party processor. It replaces both with a service you now operate, upgrade, and patch. See what LiteLLM is and the head-to-head.
Requesty: the closest like-for-like
A hosted router at https://router.requesty.ai/v1 advertising 600+ models across 30+ providers and 70,000+ developers. Its pricing is arguably cleaner than OpenRouter's: a flat 5% markup on model cost, explicitly worked as "a model that costs $10 per 1M tokens from OpenAI costs $10.50 through Requesty", with no subscription, no seat fees, and no minimum spend. Free tier is 200 requests a day on free models. It also ships a native Anthropic-wire endpoint and documented Claude Code setup, which OpenRouter matches but few others do. Detail in the Requesty review.
Portkey: gateway with governance attached
Portkey is the option when the blocker is a policy question rather than a price. Guardrails, PII controls, virtual keys, budgets, and a fully MIT-licensed gateway (12,766 stars) you can run yourself. Cloud pricing is log-volume based: free for 10k logs a month with 3-day retention, $49/mo for 100k logs with 30-day retention, $9 per additional 100k, enterprise for 10M+ with VPC deployment, SOC 2 Type 2, GDPR, and HIPAA.
Helicone: observability that grew a gateway
Helicone started as LLM logging and added a gateway at https://ai-gateway.helicone.ai. If the reason you want to leave OpenRouter is that you cannot see sessions, per-user cost, or prompt versions, Helicone is a better fit than another router. Hobby is free with 10,000 requests; Pro is $79/mo; Team is $799/mo. One caveat covered in the review: the standalone self-hostable gateway repo has not shipped a functional commit since July 2025.
Vercel and Cloudflare: free, if you already live there
Vercel AI Gateway is the genuinely zero-fee hosted option: "AI Gateway charges no markup and no platform fee on tokens", including on BYOK. It meters some add-ons instead ($0.10 per 1,000 requests for a team-wide provider allowlist or team-wide ZDR, $0.075 per 1,000 custom-reporting writes). Cloudflare AI Gateway gives dashboard analytics, caching, and rate limiting free on any Workers plan, with 100,000 logs on Free and 10,000,000 per gateway on Paid; only its optional unified billing adds 5%. Both are excellent if your stack is already on that platform and unremarkable if it is not.
Direct provider keys
The alternative nobody lists because it is not a product. If you call one model from one vendor, a gateway costs you a fee, a hop, a processor, and a normalization layer, in exchange for a convenience you are not using. Two vendors is arguable. Three is where a gateway starts paying.
Coding agents are a different market
Most gateway comparisons assume you are building an application that calls an LLM. If what you are actually doing is running Claude Code, Codex, or Cursor all day, the shape of the problem changes: the client is fixed, the wire format is Anthropic's or OpenAI's, usage is bursty, and the thing you want to see is cost per repository rather than cost per endpoint.
The decision table
| If this is you | Pick | Because |
|---|---|---|
| Prototyping across many models | OpenRouter | Nothing else has the catalogue or the friction cost |
| Shipping a model-picker product | OpenRouter or Requesty | One integration, hundreds of options |
| Fee-sensitive, already on Vercel | Vercel AI Gateway | Literally 0% on tokens and on BYOK |
| Fee-sensitive, want zero third parties | LiteLLM self-hosted | No fee and no processor, at the cost of ops |
| Regulated, need guardrails and audit | Portkey or LiteLLM Enterprise | Governance is the product, not an add-on |
| Cannot answer "what did this cost per user" | Helicone | Observability first, routing second |
| One model, high volume | Direct provider keys | A gateway is pure overhead here |
| Running coding agents all day | Continuum, or BYOK direct | Flat plan beats a meter on steady agent usage |
Questions people ask
What is the best alternative to OpenRouter?
It depends on why you are leaving. For the same product with simpler pricing, Requesty (flat 5% markup, no minimum, no seats). To remove the fee and the third party entirely, LiteLLM self-hosted. For genuinely zero fees on hosted infrastructure, Vercel AI Gateway. For guardrails and governance, Portkey. For observability, Helicone.
Is there a free alternative to OpenRouter?
Several. LiteLLM and Portkey both ship fully open-source gateways you can self-host at no licence cost. Vercel AI Gateway charges no markup and no platform fee on tokens, including with your own keys. Cloudflare AI Gateway is free for analytics, caching, and rate limiting. In every case you still pay the model provider.
Is Requesty cheaper than OpenRouter?
Roughly the same, with different edges. Requesty takes a flat 5% on inference with no minimum and no subscription; OpenRouter takes 5.5% on credit purchases with a $0.80 minimum. Requesty wins on small top-ups. OpenRouter wins if you use BYOK under its $25,000 monthly allowance, where its fee is zero.
Should I self-host a gateway instead of using OpenRouter?
Only if you have the operational capacity. Self-hosting LiteLLM or Portkey removes the fee and the third-party processor, but you inherit patching, upgrades, and availability. LiteLLM shipped 15 release tags in the 12 days to 19 August 2026 and disclosed several critical CVEs in April 2026, which is the workload you are taking on.
Does OpenRouter have a competitor with no fees at all?
Vercel AI Gateway is the clearest one: its docs state it charges no markup and no platform fee on tokens, and no fee on bring-your-own-key either. It meters some optional capabilities separately, such as $0.10 per 1,000 requests for a team-wide provider allowlist or team-wide zero data retention.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- OpenRouter pricing 5.5% fee, BYOK allowance
- Requesty pricing 5% markup, 200 free requests/day, no seat fees
- Portkey pricing free 10k logs, $49/mo Production, $9 per 100k overage
- Helicone pricing Hobby free, Pro $79/mo, Team $799/mo
- Vercel AI Gateway pricing no markup, no platform fee, add-on surcharges
- Cloudflare AI Gateway pricing free core, log limits, 5% unified billing fee
- LiteLLM on GitHub star count and release cadence read 19 Aug 2026