Read from vendor pricing pages on 19 August 2026: OpenRouter charges 5.5% on card credit purchases (minimum $0.80) and 5% on BYOK above a $25,000-a-month allowance. LiteLLM open source is $0 with enterprise quoted per gateway capacity. Portkey is free to 10,000 logs, $49/mo for 100,000, overage $9 per 100,000. Helicone is free to 10,000 requests, $79/mo Pro, $799/mo Team. Vercel AI Gateway takes zero markup on tokens including BYOK and meters add-ons separately. Cloudflare AI Gateway core is free with a 5% fee only on Unified Billing credits. Continuum includes the gateway, keeps BYOK free forever, and prices hosted inference from $25/mo.
- Percentage fees are irrelevant at $200/month and dominate at $50,000/month. Pick your volume first.
- Vercel takes zero markup on tokens, BYOK included. That is the outlier position.
- Portkey and Helicone bill on logs and requests, not tokens. Volume of calls, not size of calls.
- Cloudflare's core gateway is free; the 5% applies only if Cloudflare buys the tokens for you.
- "$0 open source" is a licence price, not a total. Add compute plus a share of an engineer.
- Continuum's gateway is included: BYOK free forever, hosted inference $25 to $500 a month.
The master table
Every figure below was read from the vendor's own pricing page or documentation on 19 August 2026. Gateway pricing changes often; re-check before you budget.
| Gateway | How it charges | Free tier | Paid entry | The catch |
|---|---|---|---|---|
| OpenRouter | % of credit purchases | Free models available | 5.5% on card ($0.80 min), 5% crypto | BYOK is fee-free to $25,000/mo of list-price inference, then 5% |
| LiteLLM | Free OSS, enterprise quoted | Entire OSS gateway, no request cap | Enterprise: contact sales | Priced on gateway request capacity, never per token, and not published |
| Portkey | Per recorded log | Self-hosted OSS unlimited; Developer 10,000 logs/mo | $49/mo Production, 100,000 logs | Overage $9 per additional 100,000, up to 3 million |
| Helicone | Per request plus storage | Hobby 10,000 requests/mo, 1 GB, 7-day retention | $79/mo Pro | Team is $799/mo; storage overage around $0.97/GB |
| Vercel AI Gateway | Nothing on tokens | Monthly free credits, limited model set | $0 markup; you buy credits at list | Add-ons metered: $0.10 per 1,000 requests for team-wide allowlist or ZDR |
| Cloudflare AI Gateway | Free core, fee on Unified Billing | Analytics, caching, rate limiting; 100,000 logs on Workers Free | $0 for the gateway itself | 5% on credits purchased through Unified Billing; Logpush $0.05/million over 10 million |
| Kong AI Gateway | Free OSS, Konnect subscription | Kong Gateway OSS, Apache 2.0 | Konnect Plus, around $105/mo per gateway | Konnect pricing is not published cleanly; verify with Kong before budgeting |
| Continuum | Included with the app | BYOK free forever, full app | Plus $25/mo for hosted inference | Hosted keys need a paid plan; not a standalone proxy for app traffic |
The four billing models, and who each one suits
- Percentage of spend (OpenRouter, Cloudflare Unified Billing). Scales perfectly with usage, which is delightful at small volume and brutal at large. It also has the nicest property of any model: at zero usage you pay zero.
- Per log or per request (Portkey, Helicone). Decoupled from token spend, so a workload of many tiny calls is expensive and a workload of a few enormous calls is cheap. Check your call count, not your bill, before you compare.
- Zero markup, platform-attached (Vercel). You pay provider list price and the gateway is a feature of a platform you already buy. Unbeatable if you are on it, unavailable if you are not.
- Free licence, your operations (LiteLLM, Portkey OSS, Bifrost, Kong OSS). The only model where the cost is invisible on a spreadsheet, which is exactly why it gets chosen too often.
Worked totals at three volumes
Assume the gateway is not the model bill: these are the additional costs of putting a gateway in the path. Token spend is the same in every column.
| Scenario | OpenRouter | Portkey | Helicone | Vercel | Self-hosted LiteLLM |
|---|---|---|---|---|---|
| Solo dev $200/mo tokens, 20,000 calls | $11 | $0 (Developer covers 10k, so $49 if you log all 20k) | $79 (over the 10k free requests) | $0 | $0 licence, a few dollars of compute, your evening |
| Small team $3,000/mo tokens, 400,000 calls | $165 | $49 + $27 overage = $76 | $79 | $0 | $0 licence, roughly $100 compute, some of an engineer |
| Scaling company $50,000/mo tokens, 5 million calls | $2,750 | Enterprise quote (over the 3 million overage cap) | $799 Team, plus storage | $0 on tokens, plus metered add-ons | $0 licence, real infrastructure, a named owner |
Two things fall out of that table. First, the percentage model loses badly at scale: $2,750 a month buys a lot of Postgres. Second, the log-based products are almost volume-independent in the middle band, which is why a team doing $3,000 of tokens across 400,000 small calls should look at Portkey and a team doing $3,000 across 4,000 huge calls should look at OpenRouter.
Where Continuum sits
Continuum prices differently because it is not sold as a gateway: the gateway is a component of a workbench for coding agents.
| Tier | Price | Weekly hosted allowance |
|---|---|---|
| Free | $0 | None. BYOK with your own provider keys, full app. |
| Plus | $25/mo | $25 a week of hosted usage |
| Max 100 | $100/mo | $100 a week |
| Max 200 | $200/mo | $200 a week |
| Ultra | $500/mo | $1,000 a week |
| Team | from $25 per member/mo | Same tiers on one subscription, billed by live member count |
Questions people ask
How much does OpenRouter charge?
A 5.5% fee with a $0.80 minimum on credit purchases made by card, and 5% for cryptocurrency payments, checked August 2026. Bring-your-own-key usage carries no fee up to a monthly allowance of list-price inference, documented as $25,000 a month on pay-as-you-go and $200,000 on enterprise, then 5% above that.
How much does Portkey cost?
The self-hosted open-source gateway is free with no request limit. On the hosted platform, Developer is free with 10,000 recorded logs a month, Production is $49/mo with 100,000 logs and overage at $9 per additional 100,000 up to three million, and Enterprise is custom priced at 10 million plus logs with SSO, VPC hosting, and SOC 2 Type 2.
How much does Helicone cost?
Hobby is free with 10,000 requests a month, 1 GB of storage, and 7-day retention. Pro is $79/mo with unlimited seats and one month of retention. Team is $799/mo with five organisations, SOC 2 and HIPAA, and three months of retention. Ingestion is capped by tier, from 10 logs a minute on Hobby to 15,000 on Team.
Is Vercel AI Gateway really free?
There is no markup and no platform fee on tokens, including when you bring your own provider key, so you pay provider list price through purchased credits. Some capabilities are metered separately: custom reporting at $0.075 per 1,000 tag writes and $5 per 1,000 report queries, and team-wide provider allowlist or zero data retention at $0.10 per 1,000 requests each.
Which AI gateway is cheapest?
At low volume, the free tiers: Cloudflare's core gateway, Vercel's zero markup, or Continuum's free BYOK app. In the middle band a $49 to $79 monthly plan usually beats a percentage fee. At high volume self-hosted LiteLLM wins on paper, and only if you already have the people to run it.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.