Pyvorin for Startups on a Budget

Startups do not have a compute problem; they have a runway problem. Every pound of cloud spend is a pound that cannot pay for the next hire, and every month of infrastructure headroom is a month the company gets to keep building. Pyvorin fits that reality only when it is used with startup discipline: measure your own workload first, buy the smallest published tier that fits, and walk away if the measurement says walk away. This page gives the verified pricing, a labelled illustrative runway model, and the three cases where the honest answer is not to buy.

Verified pricing

These figures come from the published plan table, checked against the billing database on 13 September 2026. They are class A facts, and they are the only pricing a startup should plan on.

Annual billing is priced at ten times the monthly figure, so paying annually saves roughly two months. Every paid plan carries a trial long enough to run real measurements — seven days on Starter and Team, fourteen on Business and Enterprise. Activation is a single command and the only network call the package makes; compilation itself is local and in-process, so no source code leaves your infrastructure at any point.

Why credits and free tiers change the maths

Accelerator credits — the cloud programmes most startups live on — are finite but fungible. A workload that occupies half the CPU time consumes credits at half the rate, and the same grant lasts proportionally longer. Whether that happens depends on the usual conversion rule: spend must scale with the accelerated workload. Batch pipelines and autoscaling dev environments qualify; a fixed staging VM does not. The same rule explains why credits should be spent on measurement first: a trial that produces a real speedup number converts credit into a durable planning fact, which outlives the grant itself.

Free tiers reward the same discipline from the other side. Request and compute quotas are ceilings; finishing each request sooner means more requests fit under the ceiling before the meter starts. For a pre-revenue API, staying inside a free tier for two extra months is a real cash saving, even if no invoice ever shows it.

A labelled illustrative model

Assumptions, stated in full:

• An early-stage company spending £1,000 per month on compute, of which £600 is elastic batch and API capacity that scales with workload, and £400 is fixed (staging, databases, tooling).
• A measured 2.0x speedup on the elastic portion — an assumption, to be replaced by the company's own bench result during the trial.
• The fixed £400 does not respond.
• Plan: Starter at £49 per month, the smallest published paid tier.
• Runway context: 18 months of runway remaining — an assumption.

The modelled saving is £251 per month. Mapped onto 18 months of runway, total spend falls from £18,000 to £13,482 — a modelled saving of £4,518, which at the original burn rate is the equivalent of roughly four and a half additional months of runway — class C arithmetic, entirely dependent on the 2.0x assumption. Had the measured speedup been the suite median of 1.35x, the elastic line would fall to about £444, the monthly total to £893, and the modelled runway extension to about two months. That sensitivity is why the trial exists: measure before the card comes out.

What the paid tiers add

The Free tier exists to make measurement cost nothing. It lets a startup run the support and bench commands, find out which functions compile at full tier, and get a real number for the hot path before any purchasing conversation. The paid tiers add what a company needs as the compiled code moves from an experiment into the product: higher usage ceilings, the longer Business and Enterprise trials, and terms that survive a procurement review. The upgrade path is deliberately boring — start where you are, move up when the fleet or the compliance requirements outgrow the tier. Nobody has ever regretted buying the small tier first and upgrading with data.

One practical note for the Free tier: compilation happens locally and in-process, and unsupported code falls back to CPython behaviour rather than failing. Evaluation on Free is therefore a measurement exercise, not a carve-down demo — the number you measure is the number the paid tier delivers on the same machine.

It is also worth saying what the licence is not: it is not a bet on future traffic. The runway model above turns positive only against measured behaviour on the current codebase. Buying a tier for the company you hope to be in a year is how budgets leak; buying it for the workload you can measure today is how they hold.

Three cases where the answer is no

The workload is not CPU-bound. If the product spends its time in string parsing, request routing or waiting on databases, the published suite says expect results at or below 1x. The Free tier costs nothing, and that is what it is worth here.

The compute bill is too small to matter. Below a few hundred pounds of elastic compute per month, even a strong measured speedup saves less than a Starter licence. The arithmetic in the model above only turns clearly positive because the elastic line is £600.

The code changes faster than the cache warms. Compilation amortises across repeated calls; a codebase rewritten daily on pivots will spend more time compiling than saving. Revisit after the architecture settles.

The trial cannot reach production-like conditions. A seven-day trial measured only against toy data produces a toy number. If the real workload cannot be exercised in the trial window, extend the evaluation or measure on a representative slice — a decision taken on a placeholder figure is how startups end up paying for a licence that saves nothing.

Growing into the tiers

Startups rarely stay the size they were at purchase, and the tier structure is built for that. A team that begins on Starter during evaluation moves to Team when the compiled paths become load-bearing, and to Business when procurement, longer trials or larger estates enter the picture. The annual figures — £490, £1,290 and £3,990 — are each ten times the monthly price, so the effective discount for committing is two months a year. The discipline is to upgrade on evidence rather than anticipation: a tier bought for growth that has not happened yet is a subscription to a hypothesis.

The same logic applies in reverse. If a re-architecture moves the hot path into a compiled extension language and the measured Pyvorin result falls to break-even, downgrading or cancelling is the correct financial move, and no measurement-based relationship with a vendor should make that awkward. The model on this page exists to be rerun, not admired.

Practical sequence for a budget

1. Start on the Free tier. Run python -m pyvorin support against the main entrypoints to see which functions compile at full tier.
2. Benchmark the hot path with python -m pyvorin bench --function your_entrypoint --runs 5; keep the JSON output as the record.
3. During a paid trial, rerun the bench under production-like load and watch the fallback counters on the real service.
4. Buy the smallest tier that covers the compiled functions you actually depend on. Upgrade when the fleet outgrows it, not before.

None of this requires a finance committee. It requires an afternoon and a little honesty about what the measurement said.

That honesty has a compounding effect worth naming. Startups that build a habit of measuring before buying tend to make better infrastructure decisions everywhere else — the same discipline applies to cache layers, queue systems and database upgrades. The trial is cheap tuition for a skill the company will use long after the licence decision is forgotten.

Where to go next

• Quick start — from install to a first measured result.
• How to benchmark a function — the measurement sequence in detail.
• Activation and entitlement — what activation sends, stores and checks.
• Pyvorin vs CPython: a total-cost view — the same economics at company scale.

Last reviewed 24 March 2026. Pricing is class A from the published plan table verified on 13 September 2026 (Free £0, Starter £49/mo or £490/yr, Team £129/mo or £1,290/yr, Business £399/mo or £3,990/yr, Enterprise custom; 7- or 14-day trials). All savings and runway figures are class C from the labelled illustrative model above.