The wedge claim: NemoRouter is the only LLM gateway that gives every customer all enterprise features — guardrails, A/B tests, prompt management, evals, budgets — free for life, with every major LLM provider behind one API key. Tiers vary the platform fee (4% / 2% / 0%); they never lock features.

Every LLM-gateway pricing page has the same shape: a "free" or PAYG plan with a percentage skim, one or two paid plans with a monthly minimum, and a "contact sales" door for the annual conversation. The advice that gets repeated in staff-engineering Slack is "stay on PAYG until you're sure, then commit." The advice is usually wrong — or, more precisely, it stops being right at a lower number than most teams realize.

This post draws the line. On NemoRouter, Tier 1 is PAYG at a 4% platform fee, Tier 2 is $100/month minimum at 2%, Tier 3 is $1,200/year prepay at 0% platform fee with 1,000 RPM / 1M TPM throughput. We will show, with sourced math, the exact monthly spend at which each tier becomes the cheapest option, why the same logic that picks Tier 3 also funds the provider reservations that compound the next savings round, and the three operational guards that decide whether you should still wait. ICP 2 ($2.5k–$50k/mo of LLM spend) is the audience; ICP 1 above $2.5k/mo also clears.

Five-minute path: the three tiers in one table, the breakevens, the worked example.

The three tiers in one table

Source of truth is the mono-repo pricing surface. No invented numbers, no promo "0% for life" framing — the anchor copy is "4% pay-as-you-go, 2% on Tier 2 monthly, 0% on Tier 3 annual".

Tier	Floor	Platform fee on provider spend	RPM	TPM	What you commit to
Tier 1 — PAYG	$0	4%	500	500K	Nothing. Cancel anytime.
Tier 2	$100/mo minimum	2%	500	500K	One month at a time.
Tier 3	$1,200/yr prepay	0%	1,000	1M	One year. Annual prepay funds annual provider reservations.
Enterprise	Custom	0%	Custom	Custom	Custom contract.

The feature set is identical across all four. No tier unlocks guardrails, A/B routing, prompt management, evals, per-team budgets, RLS multi-tenancy, or virtual keys — those ship on Tier 1 the moment you sign up. The only things that vary are (a) the percentage we take on top of provider cost and (b) the throughput ceilings.

That is the entire pricing surface. Compare to the four other gateways teams shortlist in 2026 — the feature-gating audit walks the same exercise across OpenRouter, Portkey, LiteLLM, and Helicone, all of which gate at least three governance features behind paid tiers or sales calls.

The three breakevens

There are three numbers worth memorizing. Each is the monthly LLM spend at which a tier transition is worth taking on a pure platform-fee basis. The math is grade-school arithmetic — annualize, multiply, subtract — and we encourage you to redo it on your own invoice.

Breakeven A — Tier 1 → Tier 2 at $5,000/month

Tier 2 costs $100/month, charges 2% on provider spend, and Tier 1 charges 4% with no fixed cost. Tier 2 is cheaper when the 2 percentage points you save on Tier 1's fee exceed Tier 2's $100/month floor:

0.02 × monthly_spend = 100
monthly_spend        = 5,000

Anything above $5,000/month of provider spend, Tier 2 beats Tier 1 on platform fees alone. Below $5,000/month, Tier 1 PAYG is correct.

Breakeven B — Tier 2 → Tier 3 at any spend with annual commit

Tier 3 is $1,200/year prepay at 0%. Tier 2 is $100/month ($1,200/year) at 2%. The prepay floor is identical. The flip is whether the 2% you save on Tier 2's fee covers Tier 2's running floor — which it does the moment you commit to a year:

Tier 2 annual cost  =  1,200  +  0.02 × annual_spend
Tier 3 annual cost  =  1,200            ← flat
Tier 3 wins when    :  0.02 × annual_spend > 0
                    →  always, if you can commit annually

In other words, the moment you are spending anything at all and can prepay annually, Tier 3 is the cheaper option of the two paid tiers. Tier 2 exists for teams that genuinely cannot commit annually — finance gate, procurement cycle, runway uncertainty. If those are not your constraints, Tier 3 is strictly dominant once you have cleared Breakeven A.

Breakeven C — Tier 1 → Tier 3 at $2,500/month

This is the headline number most teams want. Tier 3 wins over Tier 1 PAYG when the 4% you save covers the $1,200/year prepay:

0.04 × annual_spend = 1,200
annual_spend        = 30,000        →  $2,500/month

At $2,500/month and above ($30k/yr), prepaying $1,200 saves you the $1,200+ you would otherwise have paid as Tier 1's 4% platform fee — and you also get 2× throughput (1,000 RPM / 1M TPM vs. 500/500K) thrown in. The same throughput on Tier 2 would require a custom contract.

The breakeven table, then:

Today's monthly LLM spend	Cheapest tier on platform fee	Why
Under $2,500/mo	Tier 1 PAYG	The 4% on annual spend is less than $1,200; not worth prepaying.
$2,500/mo – $5,000/mo	Tier 3 prepay	4% saved already exceeds the $1,200 floor; Tier 2 floor still bites.
$5,000/mo – ~$10,000/mo	Tier 3 prepay	Strictly dominant over both Tier 1 (4% > $1,200) and Tier 2 (which charges $1,200 + 2%).
Over $10,000/mo	Tier 3 prepay (or Enterprise)	Tier 3 still wins; Enterprise becomes the conversation when throughput exceeds 1,000 RPM / 1M TPM sustained.

The math is the same one the cost teardown draws as Lever 2 of the four-lever 60% recipe; this post zooms in to the tier decision itself.

Worked example — the month the CFO asks

A real-shaped scenario, anonymized as acct_b3c1. Mid-market SaaS, ~80 employees, ChatGPT-shaped product feature, twelve months on Tier 1 PAYG.

Twelve-month rollup, Tier 1:

Provider spend (OpenAI + Anthropic direct, routed through NemoRouter): $84,000
Platform fee at 4%: $3,360
Gateway total: $3,360
Tier 1 floor: $0
Total annual gateway cost: $3,360

Tier 3 counterfactual, same year:

Provider spend: $84,000 (unchanged — provider PAYG retail does not change)
Platform fee at 0%: $0
Tier 3 annual prepay: $1,200
Total annual gateway cost: $1,200

Delta: $2,160/year (64% reduction on gateway cost). Plus 2× the throughput ceiling (1,000 RPM / 1M TPM vs. 500/500K), which a launch-week incident makes useful approximately once a year.

Three things this example does not claim:

It does not save 64% on the LLM bill. The provider bill is $84k either way; you are saving on the gateway layer. The 60% headline in the cost teardown requires stacking the tier change with the other three levers (gateway switch, provider reservations, prompt slop deletion).
It does not assume any reservation discount. Reservations engage post-$10k ARR on the gateway side — see the next section. The Tier 3 customer sees provider PAYG retail today; the spread becomes our margin when we cross the trigger.
It is not a promo. "0% on Tier 3 annual" is the published price, on the published pricing page, with no end date. There is no lifetime guarantee attached either — the pricing page is the contract.

If your own invoice clears the $2,500/mo Breakeven C, copy the table above into a slide and run it past your CFO. If it does not, stay on Tier 1 — that is the audit working as intended.

Why prepay is the strategy, not a discount

Most "annual contract" pages are a flat discount: pay early, save 10–20%, the vendor's margin is the same either way. NemoRouter's Tier 3 is structurally different, and the structure is the wedge.

The mechanic: after $10k ARR, aggregated customer volume becomes large enough to buy provider-side reservations — Azure OpenAI PTU, Google GSU / Committed Use Discounts, AWS Bedrock Provisioned Throughput. Yearly reservations save up to 70%, monthly up to 30%. Customers continue paying full PAYG retail through the gateway; the spread becomes our gross margin.

Tier 3 is the funding instrument for that mechanic. Annual prepay from the customer arrives once, in cash, on day one of a twelve-month term. We deploy a portion of that into provider reservations whose annual savings shape match the customer's annual commitment shape. The customer's price does not change — they continue paying provider PAYG retail through the gateway — and the reservation spread becomes our margin instead of being clawed back through a higher platform fee.

The published reservation economics, sourced from each provider's own pricing surface (linked in Sources):

Provider	Reservation product	Annual discount vs. PAYG	Monthly discount
Microsoft Azure OpenAI	Provisioned Throughput Units (PTU)	up to 70%	up to 30%
Google Cloud Vertex AI	Provisioned Throughput + 1y/3y Committed Use Discounts (CUDs)	~37% (1y) / ~55% (3y)	n/a
AWS Bedrock	Provisioned Throughput (model units)	~30–50%	hourly + monthly options

Those numbers belong to the provider, not to NemoRouter. We do not promise the customer a 70% discount; we promise a 0% platform fee on Tier 3 today, and we use the provider spread to fund the business that delivers it.

When NOT to prepay — three operational guards

We would rather you stay on Tier 1 than prepay against the wrong shape. Three cases where Tier 3 is the wrong choice even above $2,500/mo:

Your spend is spiky on a fortnightly horizon. If three of the last six months were below $1,000/mo and three were above $4,000/mo, the average clears Breakeven C but the trough does not. Stay on Tier 1 until the trough also clears, or hedge by buying Tier 3 the month after a high month rather than a low one.
You are evaluating a model migration that will halve your spend. If a Claude → cheaper-model swap or a Haiku 4.5 substitution is queued for next quarter, the post-migration breakeven may slip below $2,500/mo. Prepaying $1,200 against a $1,000/mo future spend is a $720/yr unforced error. Run the buyer's guide § 90-minute evaluation first, decide the model shape, then revisit the tier.
You have under 60 days of cash runway. Annual prepay is a one-time outflow. If twelve months of runway is not certain, Tier 1's $0 floor keeps your gateway bill correlated with your usage. This is the only case where we routinely tell a prospect to stay on Tier 1 even at $8k/mo of spend.

If none of those three guards trip, the math is clear. The signup form is /signup; the prepay path is /pricing.

Switching cost: zero engineering work

This is the part most teams misprice. Moving from Tier 1 to Tier 3 is a billing change, not an integration change. Same base_url, same API keys, same SDK, same routing config. The two-line OpenAI / Anthropic SDK diff from the feature-gating audit § Switching cost is migration into NemoRouter from a competitor gateway — and even that is half a day. Moving between NemoRouter tiers is one form submission on /pricing. No deploy, no SDK upgrade, no key rotation, no downtime.

That is the part of the math that does not show up in the breakeven table. If the engineering cost of a Tier 1 → Tier 3 transition were a person-week, the breakeven would shift right by several thousand dollars. It is zero, so it does not.

The 90-second decision

Three numbers from your own invoice:

Last 12 months of provider spend through your current gateway (or sum of OpenAI + Anthropic + Vertex invoices if you are still direct). Annualize if you have less than 12 months.
Divide by 12 — that is your monthly average.
Compare against the breakeven table above.

If you are below $2,500/mo: Tier 1 is correct, no decision to make.

If you are between $2,500 and $5,000/mo: Tier 3 saves you several hundred to a couple thousand dollars a year and doubles your throughput. If your CFO has even a soft preference for annual budget shape, you are done.

If you are above $5,000/mo: Tier 3 is strictly dominant on price and throughput. There is no plan B except Enterprise — and Enterprise is the conversation when throughput exceeds 1,000 RPM / 1M TPM sustained, not before.

If you would rather we run the exercise against your invoices, that is what the 30-minute walk-through on /community is for — bring the last 90 days. We will not pitch you Tier 3 if the math says Tier 1; the audit cuts both ways.

Try it on your own numbers

We auto-grant $5 in API credits on signup, no card required. Enough to exercise routing, guardrails, A/B tests, and per-team budgets on real traffic, and to validate the math before any prepay. Tier 3 is a single form away from the same dashboard.

→ Start free at nemorouter.ai/signup — Tier 1, $5 credit, no card. Already running over $2,500/mo? Skip to /pricing and pick Tier 3 directly. Mid-market SaaS or larger? Bring 90 days of invoices to a 30-min walk-through (book through /community).

Sources

Provider reservation pricing pages, verified 2026-05-16:

Azure OpenAI Service pricing (PTU): azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/
Google Vertex AI generative AI pricing (Provisioned Throughput + CUDs): cloud.google.com/vertex-ai/generative-ai/pricing
AWS Bedrock pricing (Provisioned Throughput): aws.amazon.com/bedrock/pricing/

From PAYG to Tier 3: when annual prepay beats per-call billing on an LLM gateway