OpenRouter alternatives can save money, but only if you compare the same fee. I checked vendor pricing pages on November 3, 2026, and calculated the gateway add-on for $100, $1,000, and $10,000 in provider model spend.
The short answer: Vercel AI Gateway publishes no token or platform fee; LiteLLM’s open-source gateway has no license fee if you operate it yourself. OpenRouter Standard and Cloudflare Unified Billing add 5.5% and 5% to credit purchases. octomind hub adds 3%. Portkey and Helicone sell monthly observability plans, so compare their platform price and log limits to what you need. The cheapest option depends on whether you want a hosted model catalog, BYOK, production logs, or to run the gateway.
I build octomind; I include octomind hub because it’s one of the priced options, and apply the same math to it as to other credit gateways.
Which OpenRouter alternatives cost less?
For the provider spend alone, Vercel has the lowest published gateway fee: $0 per token and $0 platform fee. That doesn’t mean every request is free; you still pay the provider’s model rate, and optional Vercel features can add charges. Vercel AI Gateway pricing (2026)
| Option | What you pay the gateway for | Best fit | Main trade-off |
|---|---|---|---|
| Vercel AI Gateway | $0 token markup or platform fee; provider list price. $5/month free credits cover eligible models. | Apps already using Vercel or the AI SDK that want one endpoint, fallback and unified spend. Vercel AI Gateway overview (2026) | BYOK requires the paid tier and purchased Gateway credits; Vercel budgets do not cap BYOK spend. |
| LiteLLM | $0 license for the open-source gateway; add your own infrastructure and operating time. | Teams that need self-hosting, provider keys inside their infrastructure, or an air-gapped deployment. | You operate availability, upgrades, secrets and logging infrastructure; enterprise governance and support require a quote. |
| Cloudflare AI Gateway | 5% on credits bought through Unified Billing; provider token rates pass through without markup. | Cloudflare users who want one bill for supported providers. | The 5% is expressly scoped to credits bought through Unified Billing. Credit purchase is card-only. |
| OpenRouter Standard | 5.5% on credit purchases; provider list price for tokens. | Hobbyists and teams that want one hosted key and a broad model catalog without operating a proxy. | Credit fees grow with spend. Standard BYOK is free through $25,000/month in list-price inference, then 5%. |
| Portkey Production | $49/month, 100,000 recorded logs included; $9 for each additional 100,000 requests. | Teams that need retries, fallbacks, key management, traces and alerts together. | Its production plan is a fixed platform cost; confirm the inference billing terms for your provider setup separately. |
| Helicone Pro | $79/month plus usage-based charges; Hobby includes 10,000 requests and 1 GB storage. | Teams that want request logging and analysis, with a gateway and caching in the same product. | Model spend is not the whole bill: usage-based storage or request charges can add to the plan. |
| octomind hub | 3% fee when you top up prepaid credits; model calls are billed at provider price. | Users of octomind cli or any OpenAI-compatible client who want a hosted endpoint and prepaid billing. | It is a hosted service, not the self-hosted octohub proxy; top-ups add a fee. |
LiteLLM’s $0 is a license figure, not a claim that a production server costs nothing. OpenRouter’s and Cloudflare’s percentages apply to credit purchases, while Portkey and Helicone charge for platform and logging features.
What does each gateway add to $100, $1,000, or $10,000 of model spend?
At the published rates, octomind hub’s 3% is the smallest percentage fee in this table, and Cloudflare’s 5% is the smallest among the third-party credit options. The table shows only gateway charges on top of the same provider spend; it excludes model tokens, tax, optional features, and unpriced infrastructure.
| Gateway fee only | $100 provider spend | $1,000 provider spend | $10,000 provider spend |
|---|---|---|---|
| OpenRouter Standard, 5.5% | $5.50 | $55 | $550 |
| Cloudflare Unified Billing, 5% | $5 | $50 | $500 |
| Vercel AI Gateway, no token/platform fee | $0 | $0 | $0 |
| octomind hub prepaid top-up, 3% | $3 | $30 | $300 |
The formula is provider spend × published credit fee. For example, $1,000 × 0.055 = $55 in OpenRouter fees on Standard. OpenRouter’s fee update lists a $0.80 minimum for non-crypto purchases; it doesn’t change these three examples, but it matters for small top-ups. OpenRouter fee update (2026)
Two other alternatives don’t fit that percentage comparison. Portkey Production’s published platform fee is $49/month with 100,000 logs included, then $9 per additional 100,000 requests. Helicone Pro is $79/month and usage-based. LiteLLM OSS costs $0 in license fees, but you need to add your actual server, database, monitoring and operator time. Portkey pricing (2026), Helicone pricing (2026), LiteLLM pricing (2026)
When is OpenRouter still the right choice?
Keep OpenRouter when one hosted key and broad model access save more engineering time than the credit fee costs. On Standard, a $1,000 monthly top-up adds $55; the page also lists automatic routing and fallbacks, and says errored attempts aren’t charged for model tokens. OpenRouter pricing (2026)
If you already hold provider keys, compare OpenRouter BYOK before switching. On Standard and Business, the first $25,000 a month of BYOK inference, counted at list price, carries no OpenRouter fee; above that allowance the fee is 5%. That’s different from the 5.5% Standard credit-purchase fee. Older articles that say BYOK becomes paid after one million requests are stale: OpenRouter’s announcement says the threshold changed in August 2026. OpenRouter pricing (2026), OpenRouter BYOK update (2026)
Choose a direct provider API if you use one model and don’t need a gateway’s catalog, shared billing, or fallback. There’s no gateway fee to compare in that setup, but you own provider-specific client code and billing. For the two-way setup choice, see LiteLLM vs OpenRouter.
Which alternative is the best LLM gateway for a hobby project, team, or compliance requirement?
Pick by the job the gateway performs, then check whether the fee base matches how you pay:
| Situation | Shortlist | Decision rule |
|---|---|---|
| Hobby or model testing | OpenRouter, Vercel free tier | Choose OpenRouter for its hosted catalog; choose Vercel if your model is eligible for the $5 monthly credit. Vercel’s free tier has a subset of models and lower per-model rate limits. |
| Startup that wants one bill | Cloudflare Unified Billing, octomind hub | Compare Cloudflare’s 5% and octomind hub’s 3% top-up fees, then check that the providers you need are supported. |
| Team with provider keys | Vercel BYOK, OpenRouter BYOK, LiteLLM | Vercel publishes no BYOK fee; OpenRouter’s allowance is $25,000/month of list-price inference; LiteLLM keeps the proxy and provider credentials in infrastructure you operate. |
| Logs, retries and fallback | Portkey, Helicone, Vercel | Compare included log volume and retention as well as the token fee. Portkey lists 100,000 logs/month at $49; Helicone lists 10,000 free requests and $79 Pro. |
| Compliance or data residency | Self-hosted LiteLLM; evaluate Portkey Enterprise | Use LiteLLM if your team can operate the gateway in its environment; Portkey lists private-cloud and VPC hosting under custom enterprise pricing. Ask vendors to confirm your required region, retention, and contract terms. |
octohub is a separate self-hosted option: it’s a Rust LLM proxy with an OpenAI-style API that routes configured aliases to provider:model targets and records requests and responses in the operator’s database. The documented quick start uses a local TOML file and SQLite. The sources I checked don’t state a self-hosted software fee, so I’m not treating it as “free.”
If the problem is model markup, compare direct keys and top-up fees first; BYOK AI: bring your own API keys, no model markup explains the key trade. If you’re choosing a provider model rather than a gateway, use the LLM API pricing comparison.
What should you verify before switching?
Before moving production traffic, check three things in the vendor’s current docs: the exact fee basis, where prompts and keys travel, and what happens when a provider errors. A top-up fee isn’t a token markup; a zero-markup gateway can still charge for optional reporting, payment processing, storage, or an enterprise plan. Vercel AI Gateway pricing (2026), Helicone pricing (2026)
Run a small parallel sample before changing your base URL. Compare the same model and provider, prompt, output limit and retry policy. Record the provider’s invoice cost, gateway fee, fallback count, latency and error rate. Keep a direct provider key path available until those numbers match your workload. If you need the broad fee comparison again, use the equation above with your own provider spend and the fee shown on the vendor page you checked that month.
