OpenRouter Alternatives (2026): 7 Gateways Compared on Fees, Keys, Security and Logs

OpenRouter's pitch is one key for hundreds of models, and it delivers on it. The cost is a 5.5% fee on every credit purchase (minimum $0.80), a 5% fee on bring-your-own-key usage past $25,000 a month, and a service that routes and logs but does not look at what passes through. If one of those is why you are here, this is the list: seven alternatives, their real fees as published on 18 September 2026, what each one does to a request, and a decision at the end.

Why people leave OpenRouter

Not everyone should. If you want the widest model catalog under one key and you spend little, 5.5% is a rounding error and OpenRouter is fine. The reasons that actually send people looking:

• The fee compounds. At $2,000 a month in tokens, 5.5% is $110 a month for a router. Two of the alternatives charge nothing for the same job.
• Nothing is checked. An agent that reads web pages, e-mail or documents can be handed instructions by them. OpenRouter forwards the request as-is. It was never meant to be a security layer, and it does not claim to be.
• The log is a debug log. You can see your requests. You cannot hand a client a record that proves what was sent, because the log can be edited by whoever holds the account.
• It cannot sit in front of a subscription. Claude Code on a Max plan, Codex on ChatGPT Pro: OpenRouter needs an API key, so it cannot be the address those tools send to without switching you to per-token billing.
• Data residency. If you need traffic kept in the EU, you need to read the small print of every router on this list, and some make it easier than others.

Match your reason to the table, then read the two or three sections that apply.

The comparison

Fees are what the vendor's own pricing page said on 18 September 2026. "Fee on top" means the amount added to the model provider's list price when you pay through the gateway. BYOK means bringing your own OpenAI or Anthropic key.

Gateway	Fee on top of provider price	BYOK	Screens for injection	Audit log	Spend caps	Free tier
OpenRouter	5.5% on credits (min $0.80); 5% crypto	Free to $25k/mo, then 5%	No	Debug log	Per key	Free models only
Constellation Gate AI	0% ("the same price you would pay the provider directly")	Free, keeps subscriptions	Yes; flags on Free, blocks on Pro	Hash-chained, anchored, verifiable	Pro	20,000 recorded requests/mo
Vercel AI Gateway	0%, "no markup and no platform fee on tokens"; card fees apply	Free on paid tier	No	Traces (paid add-on)	Budgets	Subset of models, rate-limited
Cloudflare AI Gateway	5% on Unified Billing credits; 0% with your own keys	Free	Content-safety guardrails (billed as Workers AI); DLP free	Logs, 100k free / 10M paid	Rate limits	Core features free
Requesty	5% ("$10 from OpenAI costs $10.50 through Requesty")	Yes	PII masking; no injection claim	Dashboards	Per user	200 requests/day on free models
Portkey	None stated; plan fee instead	Yes	Guardrails incl. injection scan, tiered by plan	Logs, 10k free / 100k on $49	Yes	10k logs/mo
LiteLLM	0%, self-hosted; Enterprise priced annually, "never per token"	Yes	Via plugins you configure	Your database	Yes	Open source, forever
Helicone	None stated; plan fee instead	Yes	No	Logs, 10k free	Yes	10k requests
Direct keys	0%	n/a	No	Provider console	Provider limits	Provider's own

Constellation Gate AI

Gate is the one on this list that started from the security side and added the router, rather than the other way round. It calls itself "the accountability layer for AI" and sits "between your agent and the model". Two ways to pay for tokens: run models through Gate's keys pay-as-you-go, at what Gate states is the same price you would pay the provider directly, or bring your own keys and keep paying OpenAI or Anthropic as you do now. Either way, every request is screened for prompt injection on the way in, every response on the way out, and each one is fingerprinted into a hash-chained log anchored to Constellation's Digital Evidence layer, which is the difference between a log and a receipt.

Then there is the part no other router does: it sends less. Gate removes repeated file reads, duplicated context and envelope noise before forwarding, keeping the first copy of everything, so the model gets the same content and you are billed for less of it. Gate's figures: 20% or more fewer tokens per request on agent workloads, 23% lower token cost in its Rocket Resume case study on roughly $40,000 a month. Vendor numbers; we are measuring our own and publish them on 30 September.

The catch to know before you sign up: free records, Pro blocks. The free plan (20,000 recorded requests a month, basic compression, the audit trail, Gate Connect) shows you a flagged request; stopping it, redacting PII and credentials, spend caps and the full compression are Pro at $20 per user per month. And it is the only gateway here that sits in front of Claude Code, Codex or Cursor while your subscription keeps paying for the model, which is the whole trick of the Claude Code proxy guide.

Best for: anyone whose AI reads or acts on outside material, anyone who will be asked what the AI did with a client's data, and anyone running coding agents on a subscription. Affiliate link above; we earn a commission on Pro seats, and it pays for the testing, not the verdict.

Vercel AI Gateway

The cleanest fee story on the list: "AI Gateway charges no markup and no platform fee on tokens," and BYOK carries no fee either, though BYOK needs the paid tier, which means buying credits first. You pay the provider's list price plus whatever your card processor charges. Budgets per team, project, key or member are built in, which is more than OpenRouter gives you. The free tier is a subset of models with tighter rate limits, meant for trying it, not running on it.

What it is not: a security layer. There is no screening. Observability beyond the dashboard is a paid add-on (trace drains at $0.05 per 1,000 traces, custom reporting per write and query), and team-wide zero-data-retention is $0.10 per 1,000 requests. If you build on Vercel already and your problem is purely the 5.5%, this is the shortest move.

Best for: developers on Vercel who want OpenRouter's shape without the fee.

Cloudflare AI Gateway

Free for the core features: dashboard, caching, rate limiting, and 100,000 stored logs across your gateways on the free plan. Use your own provider keys and there is no fee at all; buy credits through Unified Billing and it is 5%, so a $100 top-up costs $105. Provider prices pass through without markup. Guardrails exist, but they are content-safety checks (violence, hate, sexual content) billed as Workers AI inference per token evaluated, not a prompt-injection screen; DLP scanning for sensitive data is free on all plans.

Best for: anyone already on Cloudflare who wants caching and rate limits in front of their own keys for nothing.

Requesty

Same shape as OpenRouter, lower fee: 5% on pay-as-you-go, stated plainly as "$10 from OpenAI costs $10.50 through Requesty." Routing, caching, fallbacks, 600-plus models, bring your own keys, budget caps per user, and EU data residency, which is the reason a European solo operator might pick it over OpenRouter even before the fee. PII masking is listed; prompt-injection screening is not claimed. The free tier is 200 requests a day on free models.

Best for: people who like OpenRouter's model and want half a point less fee and EU residency.

Portkey

Portkey does not add a percentage; it sells plans. Developer is free with 10,000 recorded logs a month, Production is $49 a month for 100,000 (then $9 per extra 100,000 up to 3 million), Enterprise is custom, and the open-source gateway is free to self-host. Guardrails are its strength among the routers: more than 20 deterministic checks plus LLM-based ones including a prompt-injection scan, tiered so the free plan gets the basic set and Production unlocks the partner and pro tiers. Prompt management and a playground come with it.

Best for: a small dev team that wants an observability and guardrails console and is happy paying a flat fee rather than a percentage.

LiteLLM

The open-source proxy, free to self-host forever, speaking OpenAI's format to more than a hundred providers. There is no fee because there is no one to pay; there is also no one to call, no dashboard until you deploy one, and a server that is now yours to patch. Enterprise is priced annually on request capacity and support, "never per token." Guardrails and logging are plugins you wire in. If you have read our post on what Claude Code sends through a gateway, you know a gateway has to keep up with the tools it fronts; with LiteLLM, keeping up is your job.

Best for: teams with an engineer who wants the proxy inside their own network and does not mind running it.

Helicone

Observability first, gateway second. Free to 10,000 requests and 1 GB of storage, Pro at $79 a month, Team at $799, with the gateway feature on Pro and up. No markup is published; the plan fee is the cost. Excellent if what you actually wanted from OpenRouter was the usage dashboard and you have outgrown it; not a security layer and not the cheapest route to a second model.

Best for: people whose problem is "I cannot see what my app is spending" rather than "I am paying too much for the router."

Going direct

If you use one provider, the cheapest gateway is none. Anthropic and OpenAI both give you a console with usage, spend limits and prompt caching, and nobody takes a percentage. What you lose is the one-key-many-models convenience, any screening, and a log you could show a third party. Most solo operators started here and left for a reason; if the reason was only "I wanted to try another model", a free-tier gateway is the answer, not a paid one.

The cheaper-Claude question, answered honestly

A lot of people arrive at this page searching for a cheaper Claude API. There is no such thing: every router here passes Anthropic's list price through, and the ones that charge a fee charge it on top. The money is in four other places, and a good gateway helps with all of them:

1. No markup. Vercel and Gate, 0%. That is the OpenRouter fee back in your pocket.
2. Prompt caching. Anthropic's own feature; any gateway that forwards the headers lets you keep it.
3. Fewer tokens per request. Only Gate does this among the seven: strip the repeated file reads and duplicate context before the model sees them. This is where the 20% figure comes from, and it applies whichever model you use.
4. Cheaper models for cheaper tasks. Your decision, in your config. Subagents, classification, summaries: not everything needs the frontier model.

Which one

• The fee is the problem, nothing else: Vercel AI Gateway, or Cloudflare with your own keys.
• Your AI reads or acts on things you did not write: Gate. It is the only one on the list whose first job is screening.
• You will be asked what the AI did with someone's data: Gate, for the log you cannot edit afterwards.
• You run Claude Code or Codex on a subscription and want it fronted: Gate, via Gate Connect.
• You want OpenRouter, but European and half a point cheaper: Requesty.
• You want a guardrails console and prompt management for a flat fee: Portkey.
• You want it in your own network and have someone to run it: LiteLLM.
• You want to see spend, not save it: Helicone.
• One provider, no agents, nothing sensitive: go direct.

Your first fifteen minutes

Following the rule we apply to every tool: real input, one output, hard stop.

1. Minutes 0 to 3. Open your OpenRouter activity page and write down last month's token spend and the fee paid on top. That number is the case for or against moving.
2. Minutes 3 to 8. Create a free account on the one gateway your reason points to. For most readers of this site that is Gate: no card, one key, and if you run Claude Code, Gate Connect covers it in one toggle.
3. Minutes 8 to 13. Point one tool at it, the one you used most yesterday, and do a real task through it.
4. Minutes 13 to 15. Open the dashboard. Note requests recorded, cost, and (on Gate) tokens saved. Stop.

Output: two numbers on a sticky note: what OpenRouter's fee cost you last month and what the first real task cost through the alternative. A week later, compare. If you moved for security rather than fees, the number you are waiting for is the first flagged request, and the day you see one is the day the Pro plan becomes a simple decision.

Fees and features checked on each vendor's pricing page on 18 September 2026. Gateways change their pricing more often than most software; if a number above is out of date, tell us and we will fix it with a dated note.