InfratoApp / Tools

Tools that run on the directory

Q: Are the calculators estimates?

No. Every rate is a figure read from a seller’s own published page and stored with the date it was read, and the arithmetic is the seller’s own pricing rule. What the tools cannot know is your discount, your committed-use terms or your egress — so the answer is the list price for the workload you described, not a quote.

Q: Why is every tool deterministic?

Because a number you cannot reproduce is not evidence. The same inputs give the same answer on every run, the URL carries your inputs, and each figure links to the page it was read from — so somebody else can check the result rather than take it on trust.

Q: What happens where you hold no data?

The gap stays a gap. A tool will say it cannot answer rather than fill the hole with a vendor’s marketing claim or an average of unrelated numbers.

Every tool runs on the same sourced data as the directory — same inputs, same answer, every time. Deterministic, cited, shareable by URL.

Built and running

LLM cost calculator

Your token mix, cached share included, ranked across every seller.

200,000,000 in · 50,000,000 out / mo

meta-llama/llama-3.1-8b-instruct · Novita AI$6.50

mistralai/mistral-nemo · OpenRouter$7.30

inclusionai/ling-3.0-flash · OpenRouter$7.35

Open with these numbers →

/llm-cost · rates read < 15h ago

GPU cost calculator

1× H100-SXM-80GB · 24/7

730 hours a month

TensorDock$1,642.50/mo

Vast.ai$1,752.00/mo

Thunder Compute$2,153.50/mo

Open with these numbers →

/gpu-cost · rates read < 15h ago

Stack builder

A few questions → a complete stack with per-pick reasoning and a build order.

It names the layers you actually need and the ones you can skip, and says where we hold too little to answer rather than guessing on your behalf.

Build a stack →

Compare

Two products, two GPUs or two models, attribute by attribute.

Every row cites its source, and no row declares a winner where one side has no answer on record.

Open the comparison hub →

Planned

not built

Open vs closed weights

Ships when hosted rates and GPU rates share one freshness window — comparing a price read today against one read last week would answer the wrong question.

Price-drop alerts

Needs a daily capture we can vouch for on every seller, not most of them.

Planned tools are listed rather than hidden. A roadmap tells a reader what this is becoming, and naming what each one waits on is a claim they can check.

How these tools work

Live rates, not estimates: Each figure is read from the seller’s own page and stored with the date it was read. Anything older than 21 days is hidden rather than shown stale.
Share exactly what you built: Your inputs live in the URL, so the link you send resolves to the same answer you saw.
Gaps stay gaps: Where we hold nothing, a tool says so. It does not average unrelated numbers into a figure that looks like an answer.

Frequently asked

Are the calculators estimates?

No. Every rate is a figure read from a seller’s own published page and stored with the date it was read, and the arithmetic is the seller’s own pricing rule. What the tools cannot know is your discount, your committed-use terms or your egress — so the answer is the list price for the workload you described, not a quote.

Why is every tool deterministic?

Because a number you cannot reproduce is not evidence. The same inputs give the same answer on every run, the URL carries your inputs, and each figure links to the page it was read from — so somebody else can check the result rather than take it on trust.

What happens where you hold no data?

The gap stays a gap. A tool will say it cannot answer rather than fill the hole with a vendor’s marketing claim or an average of unrelated numbers.

1116 products · 2,416 sourced answers · How we read this data →