Flat-fee BYOK vs percentage-fee aggregators: what each actually charges (2026)
The common framing — BYOK gateways charge no markup, aggregators resell tokens at a premium — is wrong, and this page used to repeat it. OpenRouter's own documentation states it passes provider pricing through without any markup on inference. Both models charge $0 on tokens. The real difference is the shape of the platform fee: a flat monthly subscription against a percentage of what you top up.
BYOK vs managed LLM providers — at a glance
| Dimension | Flat-fee BYOK gateway | Percentage-fee aggregator |
|---|---|---|
| Markup on tokens | None | None — provider pricing passed through |
| How the platform charges | Flat $0 / $49 / $499 per month | 5.5% on card credit purchases (5% crypto, $0.80 min) |
| Cheaper below ~$891/mo token spend | No | Yes |
| Cheaper above ~$891/mo token spend | Yes | No |
| BYOK supported | Yes, always, no per-request fee | Yes — 1M requests/month free, then 5% of equivalent platform cost |
| Who holds the provider contract | You | You (BYOK) or OpenRouter (credits) |
| Provider discounts reach you directly | Yes — batch 50%, Anthropic cache reads 0.1x, Google cached input 90% off | Yes on BYOK; mediated on credits |
| Provider sign-ups required | Yes, one per provider | No, on credits |
| Best for | Above ~$891/mo, regulated teams, predictable billing | Below ~$891/mo, prototypes, breadth of models without sign-ups |
Pick BYOK gateways (e.g., VerticalAPI) or Managed aggregators (e.g., OpenRouter)?
When a flat-fee BYOK gateway wins
A flat platform fee stops scaling with your bill. VerticalAPI charges $0 / $49 / $499 per month and nothing on tokens — you hold the provider accounts and they invoice you directly, so provider-side discounts (Anthropic batch at 50%, cache reads at 0.1x input, Google batch at 50% and cached input at 90% off) land in your account rather than being mediated. Above roughly $891/month of token spend the flat $49 is cheaper than a 5.5% fee on the same volume; below it, it is not.
- $0 markup on tokens; flat $0 / $49 / $499 per month
- Cheaper than a 5.5% fee above ~$891/month of token spend
- You hold the provider contracts, so batch and caching discounts are yours directly
- Enforced rate limits: 10/min free, 60/min Pro, 300/min Enterprise
When a percentage-fee aggregator wins
A percentage fee costs almost nothing at low volume, and OpenRouter charges no markup on inference either — its own documentation states it passes provider pricing through unchanged. What you pay is 5.5% when buying credits by card (5% by crypto, $0.80 minimum). Below roughly $891/month of token spend that is less than a $49 flat fee, and you skip signing up with each provider. OpenRouter also supports BYOK, with the first 1M requests per month free and 5% of the equivalent platform cost after that.
- $0 markup on inference — provider pricing passed through unchanged
- 5.5% fee on card credit purchases (5% crypto, $0.80 minimum)
- Cheaper than a $49 flat fee below ~$891/month of token spend
- BYOK supported: first 1M requests/month free, then 5% of equivalent platform cost
Evaluate on one, run production on the other
The two are not mutually exclusive. Use an aggregator's credits to try models without signing up everywhere, then move the workloads that carry volume to a flat-fee BYOK gateway once spend passes the break-even. Both speak the OpenAI shape, so it is a base-URL change.
from openai import OpenAI # evaluation: aggregator credits, no provider sign-ups evalc = OpenAI(base_url="https://openrouter.ai/api/v1", api_key="sk-or-...") # production: flat fee, your own provider key, no token markup prod = OpenAI(base_url="https://api.verticalapi.com/v1", api_key="vapi_...") resp = prod.chat.completions.create( model="claude-sonnet-5", messages=[{"role": "user", "content": "Hello"}], extra_headers={"X-Provider-Key": "sk-ant-..."}, )
VerticalAPI verdict
Model your own token spend and compare it to $891 a month; that single number decides most of it. Below the line, a percentage fee is cheaper and skipping provider sign-ups is worth real time — OpenRouter is the better choice and this page is not going to pretend otherwise. Above the line, a flat fee stops scaling with your bill and the gap widens every month. Independently of cost, prefer BYOK when you need the provider contract in your own name for compliance, or when you want batch and prompt-caching discounts credited to your account rather than mediated. VerticalAPI is BYOK with a flat fee; OpenRouter offers both credits and BYOK, so the two are not mutually exclusive.
Frequently asked questions
Do aggregators mark up token prices?
OpenRouter does not. Its documentation states it passes the pricing of the underlying providers through without any markup on inference. Earlier versions of this page claimed a 5% token markup, which was wrong. What OpenRouter charges is a 5.5% fee when you buy credits with a card (5% for crypto, $0.80 minimum) — a fee on the amount you top up, not on the tokens you consume.
So which is actually cheaper?
Divide the flat fee by the percentage. VerticalAPI Pro is $49 a month and OpenRouter's credit fee is 5.5%, so they cost the same at $891 of monthly token spend. Below that, the percentage fee is cheaper: at $200 a month of tokens it is $11 against $49. Above it, the flat fee wins and the gap grows: at $3,000 a month the percentage costs $165. If your spend is well under the line, VerticalAPI's Free plan or OpenRouter credits are both reasonable and the decision is not about money.
Can I use my own provider keys with an aggregator?
With OpenRouter, yes. The first 1M requests each month are free on standard plans and 5% of the equivalent platform cost applies after that; Enterprise raises the free threshold to 5M requests. So BYOK is not a dividing line between the two products — the dividing line is whether the platform fee is flat or proportional.
What do I actually gain by holding the provider accounts myself?
Three things that are hard to get any other way. Provider-side discounts land in your account: Anthropic's Batch API is 50% off with cache reads at 0.1x of input, and Google's batch is 50% off with cached input 90% off. Compliance sits in your name — your DPA, your data-residency choice, your SOC 2 evidence chain. And spend caps are enforced at the provider, which is a harder limit than a dashboard setting.
What is the actual downside of BYOK?
Setup and operations. You sign up with each provider, hold each key, and reconcile several invoices instead of one. For a prototype or a solo project that overhead is not worth $49 a month of savings you are not yet making. It becomes worth it when token spend passes the break-even or when someone asks whose name is on the data-processing agreement.
Can I use both?
Yes, and plenty of teams do — an aggregator for breadth while evaluating models, then a flat-fee BYOK gateway for the workloads that go to production and carry the volume. VerticalAPI exposes a single OpenAI-compatible endpoint at https://api.verticalapi.com/v1, so moving a workload across is a base-URL change.
Limitations of this comparison
- Fees verified against OpenRouter's own documentation on 29 July 2026. Aggregator fee structures change more often than model prices — re-check before making a commitment on these figures.
- The $891/month break-even compares VerticalAPI Pro at $49 against a 5.5% card credit fee. Paying by crypto (5%) moves the line to $980; the Enterprise plan at $499 moves it to about $9,073.
- The break-even ignores the $0.80 credit-purchase minimum, which matters only at very low volume.
- Monthly request quotas are deliberately not compared here: VerticalAPI's advertised quotas and its enforced TIER_LIMITS do not currently agree, so only the enforced per-minute rate limits are quoted.
- This page compares fee structure, not model availability or routing quality, which are separate questions and can dominate the decision for some teams.
What may change in 12-24 months
- Aggregator pricing is converging on payment and platform fees rather than token markups, which makes 'no markup' claims true on both sides and pushes the real comparison onto fee shape.
- BYOK is becoming table stakes: OpenRouter already supports it with a generous free threshold, so it is no longer a differentiator on its own.
- Provider-side discounts are deepening — batch at 50%, cache reads at a tenth of input — which raises the value of holding the provider account directly.
- Compliance is moving from a nice-to-have to the deciding factor for regulated buyers, and it favours whoever holds the contract in the customer's own name.
Related questions
ChatGPT, Perplexity and Gemini usually suggest these next.
- At what monthly token spend does VerticalAPI Enterprise beat a 5.5% credit fee?
- Does OpenRouter's 1M free BYOK requests cover a production chatbot?
- How do Anthropic batch and prompt-caching discounts work through a BYOK gateway?
- Who signs the data-processing agreement when you use an aggregator's credits?
- Can I evaluate on an aggregator and move to BYOK without changing code?
More head-to-head provider comparisons
OpenRouter vs VerticalAPI: aggregator vs BYOK gateway
GPT-4o vs Claude Sonnet 4.5 direct head-to-head
Enterprise LLM hosting head-to-head
How the two caching models compare
Self-hosted vs serverless inference
Run a provider we have not measured? Any public endpoint is benchmarked and listed for free in the API directory. If you want yours measured now, on the record — time to first token, throughput under concurrency, cost per million tokens, function-calling reliability — see how a verified run works. The result is published exactly as it comes out.