LLM Routers and AI Gateways
Who sits between apps and model providers, and what they charge.
What it shows
Fees, bring-your-own-key terms and licences for 12 products, from each vendor's pricing and docs pages, checked October 10, 2026. Marketplaces resell tokens on one bill; hosted gateways proxy your calls, usually with your own keys; routers pick the model; open-source gateways run on your servers.
Compared
| Product | Type | Fee | Bring your own key | Open source |
|---|---|---|---|---|
| OpenRouter | Marketplace (resells tokens) | 5.5% on card credit purchases ($0.80 minimum), 5% on crypto. No markup on inference. | Free up to $25,000 a month of list-price usage on pay-as-you-go, then 5% of the equivalent OpenRouter cost. | No |
| Vercel AI Gateway | Hosted gateway | No markup and no platform fee on tokens. Payment processing fees may apply. Paid add-ons (team-wide ZDR, provider allowlist, custom reporting). | Supported on the paid tier with no markup or fee. | No (the AI SDK is open source) |
| Cloudflare AI Gateway | Hosted gateway | Core features free. 5% fee on credits bought through Unified Billing; provider prices passed through with no markup. | Yes (bring your provider keys; the default mode). | No |
| Requesty | Marketplace (resells tokens) | 5% on what your application spends. Free plan limited to free models, 200 requests a day. | Yes, on pay-as-you-go. | No |
| Portkey | Hosted gateway | Open-source gateway free to self-host. Hosted: free developer tier (10,000 logs a month), Production from $49 a month, Enterprise custom. | Yes (the default mode). | Yes, MIT-licensed gateway |
| LiteLLM | Open-source gateway | Open-source proxy is free; enterprise licence (SSO, RBAC, audit logs) priced by quote. | Yes (self-hosted; you hold the keys). | Yes (BerriAI, YC W23) |
| Kong AI Gateway | Open-source gateway | AI plugins ship free in open-source Kong Gateway; advanced plugins and the Konnect control plane are paid. | Yes (self-hosted). | Yes, Kong Gateway is Apache 2.0 |
| Helicone | Open-source gateway | Self-hosting free (Apache 2.0). Hosted: free Hobby tier, Pro $79 a month, Team $799 a month. | Yes. | Yes, Apache 2.0 |
| Martian | Model router | Not published. | Not published. | No |
| Not Diamond | Model router | Not published; custom terms through AWS Marketplace. | Yes (requests run on your keys). | No |
| Unify | Model router | Not published. | Not published. | No |
| AiHubMix | Marketplace (resells tokens) | Per-model prices on its site; terms not published in English. | Not published. | No |
Billing and funding
| Product | Who you pay | Latest funding | Sources |
|---|---|---|---|
| OpenRouter | Prepaid OpenRouter credits; OpenRouter pays the providers. | $113M Series B led by CapitalG (May 2026), after $40M seed and Series A led by a16z and Menlo (2025). | OpenRouter FAQ, OpenRouter, GlobeNewswire |
| Vercel AI Gateway | Prepaid AI Gateway credits, or invoice on Enterprise. | - | Vercel |
| Cloudflare AI Gateway | Your provider accounts, or Cloudflare credits through Unified Billing. | - | Cloudflare |
| Requesty | Prepaid Requesty credits. | GBP 2M seed led by 20VC (2025). | Requesty pricing, VCBacked |
| Portkey | Your provider accounts (keys stored in Portkey). | $15M Series A led by Elevation Capital (February 2026). | VCA, Portkey pricing |
| LiteLLM | Your provider accounts. | $1.6M seed (2023). | LiteLLM on GitHub, Caplight |
| Kong AI Gateway | Your provider accounts. | - | Kong AI Gateway, TrueFoundry |
| Helicone | Your provider accounts. | - | Helicone pricing, PricingSaaS |
| Martian | Through Martian or your own keys. | $9M seed led by NEA (November 2023); Accenture Ventures invested in 2024. | Maginative |
| Not Diamond | Your own gateway or provider accounts (it recommends a model; it does not proxy). | About $2.3M from defy.vc and angels including Jeff Dean and Ion Stoica (2024). | Not Diamond pricing, VentureBeat |
| Unify | Through Unify or your own keys. | Over GBP 6.3M seed from A.Capital, M12 and Y Combinator, among others (as of 2025). | Startups 100 |
| AiHubMix | Prepaid AiHubMix balance. | - | Mastra |
Finance your GPUs
Serving tokens on your own GPUs? Tell us the cluster, the models you serve and the contracts behind them, and we will share it with partner funders that fit.
Prefer email? hello@amcompute.com
Questions
- What is the difference between an LLM router and an AI gateway?
- A gateway is a proxy between your app and model providers: one API, your keys or its credits, plus logging, caching, rate limits and fallbacks. A router decides which model or provider should handle each request, by price, speed or predicted quality. Most products now do some of both, and marketplaces also resell the tokens.