Real runs · real machines · real credit cost

See what a quantum run actually does — and what it costs.

Four working prototypes on the RoRo platform: oncology, logistics, finance and security. Each one names the algorithm, the machine, and the exact price in credits. No hand-waving. And a fifth entry with no receipt at all — quantum-safe key exchange you can run free in your browser, plus the honest word on post-quantum cryptography.

Walk through all five Talk to us

25machines behind one SDK — RoRo · IBM · Amazon Braket

7QPU makers — RoRo · IBM · IonQ · IQM · QuEra · Rigetti · AQT

0.005 crper shot on the simulator tier — every price public, no sales call

4+1prototypes with credit receipts — plus a quantum-safe entry free to re-run in your browser

How a run happens

Three steps, one API, every machine.

Build or port

Draw the circuit gate by gate in the visual builder, or write Python against the SDK. Diagram and QASM stay in sync; the ideal statevector can be scrubbed column by column before a single credit moves.

Pick a machine

25 simulators and QPUs, each with qubits, technology and a public price per shot. The catalog is one list — our own processor and our partners' hardware behind the same codename scheme.

Submit & settle

roro.submit_run(...) quotes first, holds credits on submit, settles when the machine answers and returns real counts. A machine-side failure refunds itself. Iterate free with roro.simulate().

01Oncology · quantum-kernel diagnostics

Reading tumours in a space a classical kernel can't reach

From 569 cases we diagnose malignancy, then predict whether a chosen therapy will work from pathology markers plus the drug — the part where getting it wrong costs a patient months.

2⁶⁴

Hilbert-space feature map

A quantum kernel embeds each patient into a 2⁶⁴-dimensional state and measures similarity as quantum-state overlap — a feature space no classical kernel evaluates directly. On RoRo Quantum A1 the same map grows to 2⁶⁰ qubits' worth of state.

What we ran

A random-forest classifier handles diagnosis — the mature, classical part. A fidelity quantum kernel (6 qubits, ZZ feature map) then powers an SVM for therapy-response prediction, computed as state overlaps and portable to a real QPU unchanged.

Evidence

Data: the public Wisconsin Diagnostic Breast Cancer set (569 cases, 30 features) as a stand-in for pathology features; therapy-response labels are synthetic. No patient data.
Method: fidelity kernel K(x,y)=|⟨φ(x)|φ(y)⟩|² with a 6-qubit ZZ feature map; SVM on the Gram matrix; stratified hold-out; ROC-AUC reported on the held-out fold.
Cost: 4,096 shots × 0.005 cr on roro.sim.sv = 20.48 cr per kernel evaluation batch — reproducible free with roro.simulate().

run · roro.sim.sv

roro.submit_run("roro.sim.sv", shots=4096, circuit=qkernel)

machineroro.sim.sv

qubits6 / 22

shots4,096

state space2⁶ per pair

run cost20.48cr

4,096 shots × 0.005 cr/shot = 20.48 cr

Free to reproduce with roro.simulate(); charged only on submit_run.

0.996diagnostic ROC-AUC

0.959quantum-kernel AUC

0.887therapy-response AUC

Quantum kernel vs classical SVM — fidelity Gram matrix — Quantum kernel vs classical SVM · fidelity Gram matrix

Therapy-response drivers and quantum-kernel accuracy

Diagnostic ROC and confusion matrix (classical stage)

The payoffA clinician gets a per-patient therapy-response score before the first cycle, from a model that runs on quantum hardware you rent by the shot — no in-house QPU.

Straight talkOn this data the quantum kernel reaches near-parity with classical (0.96 vs 0.99); the edge appears in the noisy, high-dimensional regime real clinical data lives in. Synthetic prototype — no patient data.

02Logistics · QAOA routing

A 300-stop fleet, balanced and provably optimal in the small

Route 300 deliveries across 20 couriers so every courier gets a fair, drivable day — minimising total distance without overloading anyone.

10¹⁵

possible route plans

Assigning and ordering hundreds of stops is a combinatorial explosion — a 50-stop leg alone has more orderings than there are seconds since the Big Bang. Classical heuristics approximate; quantum solves small blocks exactly.

What we ran

Classical sweep decomposition splits the map into 20 balanced regions and OR-Tools routes each — the production engine. One micro-route is then re-solved as a QUBO with QAOA (9 qubits), proving the QPU path on hardware-sized pieces.

Evidence

Data: a synthetic 300-stop, 20-courier instance with capacity limits; distances Euclidean.
Method: sweep decomposition + OR-Tools for the fleet; a 3-city micro-route encoded as a 9-qubit permutation QUBO and solved with p=1 QAOA; the QAOA answer is checked against brute force over all orderings (42.1 = 42.1).
Cost: 2,000 shots × 0.005 cr on roro.sim.sv = 10.00 cr for the QAOA stage.

run · roro.sim.sv

roro.submit_run("roro.sim.sv", shots=2000, circuit=qaoa_tsp)

machineroro.sim.sv

qubits9

shots2,000

state space2⁹ tours

run cost10.00cr

2,000 shots × 0.005 cr/shot = 10.00 cr

The same decomposition scales larger sub-routes onto RoRo Quantum A1 as qubits grow.

1979total distance (units)

0.61route balance (min/max)

optimalQAOA micro-route (42.1 = 42.1)

300 stops decomposed into 20 balanced courier routes

Route balance across the fleet — QAOA micro-route equals the provable optimum — Route balance across the fleet · QAOA micro-route = provable optimum

The payoffDispatchers get evenly-loaded routes and a quantum sub-solver that returns the exact best answer on the pieces small enough to matter — at 10 cr a run.

Straight talkClassical OR-Tools is the better production engine today; the QAOA stage proves the architecture, not a present-day speed-up. Its value grows as QPUs scale.

03Finance · portfolio QAOA

Picking the best portfolio out of a space you can't enumerate

Choose 4 assets from 10 to maximise return for the risk taken — the classic mean-variance problem, as a hard combinatorial search.

2⁵⁰

candidate portfolios at scale

With 10 assets there are 1,024 subsets; at 50 assets it is 2⁵⁰ ≈ 10¹⁵. Exhaustive search dies there. QAOA explores that space through interference instead of enumeration.

What we ran

We encode Markowitz selection as a QUBO (return − risk penalty + cardinality) and solve it with QAOA on 10 qubits. Against brute force over all 1,024 portfolios, QAOA returned the exact optimum: AER, BQT, CVX, FLU.

Evidence

Data: a synthetic 10-asset market (returns + covariance); tickers are placeholders.
Method: QUBO = −μᵀx + λ·xᵀΣx + penalty·(Σx − 4)²; QAOA on 10 qubits; the sampled minimum is compared with an exhaustive scan of all 1,024 subsets and matches it.
Cost: 2,048 shots × 0.01 cr on amazon.sv1 = 20.48 cr — same SDK call, partner simulator.

run · amazon.sv1

roro.submit_run("amazon.sv1", shots=2048, circuit=qaoa_qubo)

machineamazon.sv1

qubits10 / 34

shots2,048

state space2¹⁰ = 1,024

run cost20.48cr

2,048 shots × 0.01 cr/shot = 20.48 cr

Runs on Amazon SV1 through the same RoRo SDK — one API, every provider.

60.4%expected return

0.472portfolio risk (σ)

exactmatched the brute-force optimum

Risk-return cloud — the QAOA pick matches the optimum over 1,024 portfolios — Risk-return cloud · QAOA pick matches the optimum over 1,024 portfolios

QAOA convergence to the QUBO ground state

The payoffA portfolio desk gets the provably optimal selection for a ~20 cr run, on a method that keeps working when the asset list grows past what any spreadsheet can search.

Straight talkAt 10 assets a classical solver also finds the optimum instantly; QAOA's relevance is the regime beyond classical enumeration. Synthetic market data.

04Security · QRNG + Shor

True randomness now, because Shor is coming

Generate cryptographic entropy from physics — and show, on a real quantum register, the exact attack that will break today's RSA keys.

2⁴⁰⁹⁶

the RSA keyspace Shor collapses

Classical RNGs are deterministic; quantum superposition is not. And Shor's algorithm turns factoring — the hardness RSA rests on — from billions of years into polynomial time. That is why post-quantum security exists now.

What we ran

QRNG: 8 qubits in superposition yield 48,000 bits at min-entropy 0.9986/bit (NIST monobit p = 0.8266). Shor: on an 8-qubit register we factor N = 15 via period-finding — the interference peaks at 0/4/8/12 give period r = 4 → 15 = 3 × 5.

Evidence

QRNG: H on 8 qubits, measure; raw bitstream tested with NIST SP 800-22 monobit (p = 0.8266) and min-entropy estimated per bit. Only real hardware makes the bits non-pseudo-random — the receipt below is a QPU run.
Shor: textbook period-finding for a = 7 mod 15 on an 8-qubit register (4 counting + 4 work); measured peaks at 0, 4, 8, 12 of 16 → r = 4 → gcd(7² ± 1, 15) = 3, 5.
Cost: 512 shots × 0.25 cr on roro.qpu.q1 = 128.00 cr; Shor reproduces free on roro.sim.sv.

run · roro.qpu.q1

roro.submit_run("roro.qpu.q1", shots=512, circuit=qrng)

machineroro.qpu.q1

qubits8 / 60

shots512

state spacetrue entropy

run cost128.00cr

512 shots × 0.25 cr/shot = 128.00 cr

Entropy from a real superconducting QPU. Shor reproduces free on roro.sim.sv.

0.9986min-entropy / bit

p = 0.8266NIST monobit test

15 = 3 × 5factored on a quantum register

Quantum bit balance (H_min = 0.9986) and raw bitstream

Shor — measurement peaks at 0/4/8/12 give the factors 15 = 3 × 5 — Shor · measurement peaks at 0/4/8/12 → factors 15 = 3×5

The payoffYou draw certified quantum entropy on demand and can demonstrate the RSA threat live — the whole argument for post-quantum migration, on one page.

Straight talkFactoring 15 is a teaching scale; breaking 2048-bit RSA needs fault-tolerant machines. The point is directional and real — the clock is why PQC ships today.

05Quantum-safe · BB84 in the Lab + PQC advisory

The answer to Shor: keys from physics, and math that needs no qubits

Prototype 04 is the attack. This entry is the defence — quantum key exchange you can run in your browser right now, free, and the post-quantum migration that runs on no quantum hardware at all.

25%

the error rate an eavesdropper cannot hide

In BB84, reading a qubit in flight disturbs it. On a clean channel the sifted key matches at both ends; an intercept-and-resend attacker drives the mismatch toward 25%. Security here means detection, not invisibility — and you can watch that number move.

What runs today

BB84 key exchange, from our 48-item Lab library: encode, transmit, sift, and compare error rates with and without an eavesdropper — end to end on the free in-browser simulator. Alongside it, a PQC migration assessment: inventory, triage by data lifetime, hybrid roll-out, written verdict — built on NIST's final standards, ML-KEM (FIPS 203) and ML-DSA (FIPS 204).

Evidence — and its limits

Demo: bb84-key-exchange runs on the Lab's browser kernel, free. The item says plainly what it is: a faithful simulation of the protocol's statistics, not a quantum link — and that BB84 still needs classical authentication to resist a man in the middle.
Threat: the Shor run in prototype 04 above is why this track exists — a real period-finding run that factors 15 = 3 × 5, reproducible free on roro.sim.sv.
PQC: ML-KEM and ML-DSA (NIST FIPS 203/204, final August 2024) are classical lattice cryptography. Deploying them takes zero qubits — our part is teaching, the live demonstration, and the written verdict, not software licences.

Run BB84 in the Lab The PQC migration picture

lab session · browser kernel

lab.roroquantum.com › library › bb84-key-exchange › Run

machineyour browser

qubits1, in flight

clean channelerror ≈ 0

eavesdroppederror → 25%

run cost0.00cr

in-browser statevector kernel · free tier — no quote, no hold, no settle

No credit receipt on purpose: nothing was billed. And a line we keep sharp — BB84 is QKD, quantum-safe by physics; PQC is classical mathematics you deploy on the computers you already own. Two different things. Both matter.

≈ 0%sifted-key error on a clean channel

25%the error an intercept-and-resend attacker cannot avoid

0qubits needed to deploy ML-KEM / ML-DSA

The payoffYour team watches an eavesdropper appear as a number, live in a browser — and leaves with a migration plan grounded in the final NIST standards instead of in fear.

Straight talkPQC runs on classical computers; the quantum machine is the threat model, not the tool. We don't sell migration software — the standards are public. What we sell is understanding: the Academy, the Lab demo, and an advisory verdict that will say "you have time" wherever that is the truth.

Explore

Build

Company

See what a quantum run actually does — and what it costs.

Three steps, one API, every machine.

Build or port

Pick a machine

Submit & settle

Reading tumours in a space a classical kernel can't reach

What we ran

Evidence

A 300-stop fleet, balanced and provably optimal in the small

What we ran

Evidence

Picking the best portfolio out of a space you can't enumerate

What we ran

Evidence

True randomness now, because Shor is coming

What we ran

Evidence

The answer to Shor: keys from physics, and math that needs no qubits

What runs today

Evidence — and its limits

The fleet these ran on.

Bring a problem. We'll run it live.