AI Model Router Pricing Compared 2026: OpenRouter, LiteLLM, Portkey, Cloudflare, and What You Actually Pay
A practical 2026 comparison of AI router and gateway pricing: OpenRouter, LiteLLM, Portkey, Cloudflare AI Gateway, and TheRouter. We break down token markups, platform fees, BYOK economics, and monthly cost at 1M/10M/100M tokens so you can pick the cheapest path for your workload.
TL;DR — No major AI gateway marks up tokens in 2026. The real cost difference is the platform fee on credits, the monthly subscription, and whether you can bring your own provider keys or self-host. We break down five popular routing layers below so you can compare apples to apples.
OpenAI-compatible means a provider exposes a chat-completions endpoint whose request and response shape matches the OpenAI API contract closely enough that an unmodified OpenAI SDK call works against it after swapping three values: API key, base URL, and model name. The minimum surface in practice is POST /v1/chat/completions with messages, model, and an OpenAI-shaped streaming response.
How AI Routers Charge: Three Pricing Models
Every AI routing layer sits between your code and the LLM provider. The question is how much that intermediary costs you. There are three pricing models in the wild:
- Per-token markup — a percentage added on top of the provider's per-token rate. A 10% markup on a $3/1M-token model means you pay $3.30. This compounds with every request, forever.
- Platform / credit fee — a flat percentage taken when you load credits or a monthly subscription. Predictable and independent of token volume.
- Self-hosted / open-source — you run the gateway yourself. No platform fee, but you own the infrastructure cost.
The good news: as of August 2026, none of the five gateways we compare below apply a per-token markup. Provider rates pass through at cost. The competition has moved entirely to platform fees, subscriptions, and deployment flexibility.
The Pricing Table
| Gateway | Token Markup | Platform / Credit Fee | Monthly Subscription | BYOK (Bring Your Own Keys) | Self-Host |
|---|---|---|---|---|---|
| OpenRouter | None | 5.5% on card credits ($0.80 min) | None | Yes (free up to 1M req/mo) | No |
| LiteLLM | None | None (self-hosted) | Free OSS; Enterprise by quote | Yes | Yes (MIT/Enterprise) |
| Portkey | None | None | Free tier (10k logs); Production $49/mo; Enterprise custom | Yes | Partial (gateway OSS) |
| Cloudflare AI Gateway | None | 5% on Unified Billing credits | Free (core features); Workers Paid for Logpush | Yes | No |
| TheRouter | None | Pay-as-you-go, USD balance | None | Yes | No |
Sources: OpenRouter pricing (retrieved 2026-08-25), LiteLLM pricing (retrieved 2026-08-25), Portkey pricing (retrieved 2026-08-25), Cloudflare AI Gateway pricing (retrieved 2026-08-25), TheRouter pricing (retrieved 2026-08-25).
OpenRouter: Per-Credit Fee, Massive Model Catalog
OpenRouter is the largest multi-provider aggregator by model count, offering 500+ models through a single OpenAI-compatible endpoint.
How they charge: No per-token markup. A 5.5% fee is applied when you purchase credits by card (minimum charge $0.80; crypto purchases carry a 5% fee). Provider token rates pass through at cost.
Free tier: 20+ models are available at $0 token cost. BYOK usage is free up to 1M requests per month.
What to watch: The $0.80 minimum means small top-ups are disproportionately expensive. A $10 credit purchase costs $10.80 (8% effective fee). At $100 the fee drops to the stated 5.5%.
Best fit: Developers who want the widest model selection from a single API key and don't need self-hosting or enterprise governance.
For a feature-level comparison, see our OpenRouter alternatives guide.
LiteLLM: Free Open Source, Enterprise by Quote
LiteLLM is the most popular open-source AI gateway, with 100+ provider integrations and a self-hosted proxy that you run on your own infrastructure.
How they charge: The open-source proxy is free forever (MIT-licensed for the core). No token markup, no platform fee. Enterprise adds SSO, SCIM, audit logs, secret management, multi-region control plane, and 24/7 support — priced by annual gateway request capacity, not per token.
Self-hosting cost: Free license, but you pay for the servers, database, Redis, and on-call. A minimal production setup runs $50–200/month in cloud infrastructure depending on traffic.
What to watch: Enterprise pricing is by quote only. Third-party estimates suggest plans starting around $250/month for small deployments, scaling with request volume.
Best fit: Teams that want full control over their data and infrastructure, need air-gapped deployment, or want to avoid any per-token overhead.
For a routing comparison, see our unified LLM gateway comparison.
Portkey: Observability-First with Tiered Subscriptions
Portkey positions itself as a "control panel for production AI" with strong observability, prompt management, and guardrails built into the gateway.
How they charge: No token markup. Three tiers:
- Developer (Free) — 10k recorded logs/month, 3-day log retention, 3 prompt templates. Not suited for production.
- Production ($49/month) — 100k logs/month, 30-day log retention, guardrails, RBAC. Overages at $9 per additional 100k logs.
- Enterprise (custom) — 10M+ logs, custom retention, SSO, private cloud, SOC 2 Type II, HIPAA.
What to watch: The $49/month Production plan is affordable, but log overages scale fast. A team generating 500k logs/month pays $49 + $36 = $85/month. The gateway itself is open-source for self-hosting, but managed observability is paid.
Best fit: Teams that value integrated observability, prompt management, and guardrails — and are willing to pay a predictable monthly fee rather than a per-credit percentage.
Cloudflare AI Gateway: Free Core, Pay for Extras
Cloudflare AI Gateway is unique in that its core gateway features — analytics, caching, and rate limiting — are completely free on all Cloudflare plans.
How they charge: No token markup. Core features are free. The costs come from optional add-ons:
- Unified Billing — 5% fee on credits, allowing you to pay for multiple providers through Cloudflare. Provider token rates pass through at cost.
- Logpush (Workers Paid plan) — 10M logs/month included, $0.05 per additional million.
- Guardrails — billed as Workers AI inference (token-based, using Llama Guard).
- DLP — free on all plans.
Log limits: Free plan stores 100k logs total across all gateways. Workers Paid plan stores 10M logs per gateway.
What to watch: If you skip Unified Billing and use your own provider keys, the gateway itself costs nothing. The 5% Unified Billing fee is competitive but only makes sense if you want consolidated billing through Cloudflare.
Best fit: Teams already on Cloudflare that want a zero-cost gateway with optional managed billing. Less suited for teams that need advanced routing logic or multi-provider failover beyond basic caching and rate limiting.
TheRouter: Pay-as-You-Go Routing with Fallback
TheRouter routes OpenAI-compatible requests through configured providers, supporting provider/model routing and fallback where the live product paths support it.
How they charge: Pay-as-you-go pricing against a USD balance. No separate platform fee or monthly subscription. Model pricing is listed on the models page. TheRouter provides unified billing and accounting surfaces where implemented.
What to watch: Pricing is model-specific and visible on the catalog page before you route traffic. BYOK is supported.
Best fit: Developers who want OpenAI-compatible routing with provider fallback and prefer pay-as-you-go over subscriptions.
For a migration walkthrough, see our OpenAI-to-TheRouter migration guide.
Cost Modeling: Monthly Spend at Scale
The table below estimates total gateway cost (excluding provider token charges) for three usage tiers. We assume Claude Sonnet 4.6 pricing ($3/$15 per 1M tokens) and a 50/50 input/output split.
| Monthly Tokens | OpenRouter (5.5% credit fee) | LiteLLM (self-hosted) | Portkey (Production) | Cloudflare (Unified Billing 5%) | TheRouter (pay-as-you-go) |
|---|---|---|---|---|---|
| 1M tokens | ~$0.50 | $0 (infra only) | $49/mo flat | ~$0.45 | Per-model pricing |
| 10M tokens | ~$5.00 | $0 (infra only) | $49 + log overages | ~$4.50 | Per-model pricing |
| 100M tokens | ~$50 | $0 (infra only) | $49 + log overages | ~$45 | Per-model pricing |
Notes: OpenRouter and Cloudflare fees scale linearly with spend. LiteLLM has zero gateway fees but real infrastructure costs ($50–200/month for a production proxy). Portkey's flat $49 is cheapest at high volume if you stay under the log cap. TheRouter pricing is per-model and visible on the pricing page.
When comparing API pricing across providers, always normalize to USD per million tokens and split input from output. Most providers price output tokens 2–5× higher than input tokens, so a workload heavy on completion length looks very different from a retrieval-heavy workload at the same nominal "price per million."
- Use one currency (USD) — convert at publish date and cite the rate.
- Split input/output — never quote a single blended number.
- Cite each row to the provider's own pricing page with retrieval date.
- Note context-window tiers — long-context pricing often steps higher.
Hidden Costs to Watch For
A few line items don't show up on pricing pages:
- Payment-processing fees. OpenRouter's $0.80 minimum on card purchases is effectively 8% on a $10 top-up. Cloudflare's Unified Billing passes through card processing costs.
- Log retention and overages. Portkey bills $9 per additional 100k logs. Cloudflare's free plan caps at 100k logs total. If you need long-term log storage, budget for it.
- Self-host infrastructure. "Free and open source" still costs servers, databases, and on-call. LiteLLM's zero license fee is attractive, but a production-grade deployment is $50–200+/month in cloud compute.
- Support tiers. LiteLLM Enterprise and Portkey Enterprise both charge for dedicated support. Community support is free but comes with no SLA.
Decision Guide: Pick the Cheapest Path for Your Workload
You want the lowest possible cost and own your infra: Self-host LiteLLM. Zero license fee, zero platform fee. You pay only for servers and provider tokens.
You want managed convenience with no subscription: OpenRouter (5.5% credit fee) or Cloudflare AI Gateway (free core, 5% on Unified Billing). Both pass through provider rates.
You want observability and prompt management bundled in: Portkey Production at $49/month. Predictable cost, strong dashboard, but watch log overages.
You want routing with provider fallback, pay-as-you-go: TheRouter — OpenAI-compatible routing with configured provider fallback.
You're already on Cloudflare: Start with the free AI Gateway tier. Add Unified Billing only if you want consolidated invoicing.
Frequently Asked Questions
Do AI gateways mark up token prices? Not in 2026. OpenRouter, LiteLLM, Portkey, Cloudflare, and TheRouter all pass provider token rates through without a per-token markup. They make money on platform fees, subscriptions, or self-hosted support contracts instead.
Which AI gateway is cheapest? For most teams, the cheapest path is self-hosting LiteLLM (zero gateway fees) or using BYOK on a managed gateway that waives platform fees. Among managed options, Portkey's $49/month flat fee is cheapest at high volume; OpenRouter's 5.5% is cheapest at low volume.
Is a platform fee or a token markup worse? A token markup is worse. It scales with every token you send, compounding as traffic grows. A flat platform fee on credits is a one-time percentage when you load funds — predictable and capped.
Can I use my own API keys with these gateways? Yes, all five support BYOK. With your own keys, most gateways reduce or eliminate their platform fee.
Pricing as of August 2026. Rates change; verify on each gateway's pricing page before committing budget.