Inside the platform

Six views over one corpus, dataset of 2026-09-30. The joins between them are the point: a hardware milestone that moves changes a use-case window, and a use-case requirement is what makes a milestone worth watching. Below, each view in its own shape — names, titles and dates as the platform shows them; the figures and the analysis are the subscription.

Preview · 3 / 12 · Vendors tracked

Vendor	Modality	Flagship system	Physical qubits	Logical qubits	2-qubit error	QUOPS score	QUOPS/s	Clock	Connectivity	Error correction	Next milestone	As of
Atom Computing	Neutral atom	—
Google Quantum AI	Superconducting	Willow
IBM Quantum	Superconducting	Heron r3 (ibm_boston)
▆▆▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆▆▆▆	▆▆▆	▆▆	▆▆▆▆▆	▆▆▆	▆▆▆▆	▆▆▆	▆▆▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆
▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆	▆	▆▆▆▆▆	▆▆▆▆	▆▆▆▆▆	▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆▆▆▆▆▆▆▆	▆▆▆▆▆
▆▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆▆	▆▆	▆▆▆▆▆	▆▆▆	▆▆▆▆	▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆
▆▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆	▆▆▆▆	▆	▆▆▆▆▆	▆▆▆▆	▆▆▆▆▆	▆▆▆	▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆▆▆	▆▆▆▆▆
▆▆▆▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆	▆▆	▆▆▆▆▆	▆▆▆	▆▆▆▆	▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆

The remaining rows, every figure and the analysis are part of the subscription.

Preview · 8 / 70 · Dated milestones

demonstrated projected, vendor-reported first-announced date, kept

Milestone titles and dates shown; the capacity chart, the metrics and the notes are part of the subscription.

Requirement curve · rsa2048-physical · 5 points

published estimate JoS extrapolation, labelled as ours convergence window, JoS estimate

score history usecase · sketched-streaming-fraud
2026-04-01 52 → 60 · quantum oracle sketching + shadow readout · data basis moved 2 → 3

The curve is read against the vendor roadmaps inside the platform; the vendor capacities, the use-case thresholds and the notes behind each point are part of the subscription.

Preview · 3 / 22 · Use cases assessed

Use case	Category	Industry	Total	Level	Binding constraint	Horizon	Logical qubits
Designing a nitrogen-fixation catalyst (FeMoco)	Simulation Hamiltonian simulation	chemicals, energy
Elliptic-curve cryptanalysis (secp256k1)	Cryptanalysis Discrete logarithm	security, finance
Exotic derivatives pricing	Simulation Monte Carlo simulation	finance
▆▆▆▆▆▆▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆
▆▆▆▆▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆▆▆▆
▆▆▆▆▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆▆	▆▆▆▆▆	▆▆	▆▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆
▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆▆▆
▆▆▆▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆	▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆▆▆▆	▆▆▆▆▆▆▆	▆▆▆▆▆

One complete record from the MERIQ whitepaper v1.0, as corrected in October 2026:

usecase · sketched-streaming-fraud

▶ MERIQ scorecard          V  A  D  F  R   total  level
  ───────────────────────────────────────────────────────────────────
  sketched streaming        4  3  3  2  3      60  L2 Conceptual
  fraud detection
  ▶ binding                 F  — the route has not yet run end to end
  ▶ verdict                 datable, conditionally
  ▶ horizon                 early fault-tolerant · 2029–2031

▶ score history
  2026-04-01   52 → 60   quantum oracle sketching + shadow readout
               data basis moved 2 → 3 (the route stopped needing QRAM),
               advantage and resources a rung each behind it;
               feasibility fell 3 → 2 until the route runs end to end.
               The variational card it supersedes still reads 52 — a
               new access model is a new triple, not a better score.

▶ classical baseline       gradient-boosted trees and neural nets
  performance              about 1 ms per transaction   [company-reported]
  to beat                  more fraud caught at the same false-positive
                           rate as the production ensembles
▶ value                    loss avoided · a large European card issuer
  at stake                 €10m–€100m a year            [JoS estimate]
  quantum share            none attributable yet: the proven gain is
                           memory, and a scoring stack is not memory-bound

▶ requirements             < 60 logical qubits          [published estimate]
  binding resource         logical-gate throughput, not qubit count

▶ earliest plausible window
  trapped ion              2029–2031   capacity is not the constraint —
                                       Helios already runs 48 logical;
                                       gate speed is
  superconducting          2029–2032   µs cycle times suit a stream;
                                       decoder latency is the open question

Read what the last block is doing. It is not a forecast. It is a human comparison of a published requirement against dated vendor milestones, stored with its reasoning, so you can disagree with the reasoning rather than with a number. Deriving those windows mechanically from vendor roadmaps would turn marketing into prediction, which is the one thing this dataset is built not to do.

MERIQ use-case benchmarks

Twenty-two cases: seventeen scored on the five published MERIQ criteria — value, advantage, data and I/O, feasibility, resources — and five parked at the screening gate with the reason on record. Each scored card carries its total, its level, the weakest-link cap, the binding constraint, the horizon and the earliest plausible window per technology, and is mapped to the roadmap milestones that would have to land first.

Each also names the best classical method it has to beat, with a sourced figure, and what a better answer is worth to one typical buyer, as an order of magnitude. Four of the seventeen say plainly that they are not commercially assessable today; most of the rest show money at stake with no quantum share we could defend. Writing the classical side down moved eleven scores in October 2026, our own flagship’s among them.

The remaining rows, every score and the rationales are part of the subscription.

Preview · 3 / 9 · Companies followed

Alphabet Inc.

NASDAQ:GOOGL

Metric
Revenue
R&D expense
Cash and equivalents
Liquidity
Runway [JoS estimate]

filings index earnings-call quotes commentary

Infleqtion, Inc.

NYSE:INFQ

International Business Machines Corporation

NYSE:IBM

Company names and tickers shown; the figures, the filings index, the quotes and the commentary are part of the subscription.

Change record · 2026-09 · 3 of 6 headline entries

2026-09-30milestone_newvendor:ibm●●●
Three July preprints on IBM hardware sit on the public advantage tracker as active candidates; one of their instances has since been simulated classically.
2026-09-30spec_updatevendor:quantinuum●●●
Quantinuum runs the C4-Helix code on Helios: logical Cliffords at 2.8e-4, 4.3 times below the physical gate, with no post-selection.
2026-09-30milestone_newvendor:ionq●●●
IonQ launches Superion 256, its first chip-based system: 256-qubit QPUs fabricated at SkyWater, orders open, deliveries in 2027.

Quarterly brief · 2026-Q3 · 2026-09-30

Q3 2026: one benchmark axis, three kinds of logical qubit, and a cheaper RSA

benchmarksfault-tolerancecryptanalysislistingsuse-cases

Headlines shown; the notes, the full record and the briefs are part of the subscription.

Also in the platform: Hardware basics, a referee-checked explainer of qubits, gates, error correction and seven hardware technologies with a use-case × technology matrix, glossary and simulators — readable in full on a free account.

Two add-ons, for a second question

Intelligence is the core. Two tools from our own research sit next to it on the same account and the same invoice, switched on per account on an annual subscription, on request.

Add-on · Labs · RISQ and the Shor estimator

RISQ: business risk, estimated two ways

Model a network of business risks — events, their probabilities, the dependencies between them — and estimate the tail two ways side by side: classical Monte Carlo and quantum amplitude estimation, with the same loss metrics from both: expected loss, VaR, CVaR. A sensitivity ranking orders the risk drivers by how far each one moves the result. The method is our paper, arXiv:2103.05475. The Shor estimator ships in the same module: from an RSA key size to logical and physical qubits, runtime and the roadmap year, with the error-correction sizing shown.

risq.jos-quantum.deBenchmark · precision crossover

Halving the error costs Monte Carlo four times the samples and amplitude estimation twice the queries. The platform fits the measured convergence of your own model and locates where the two cross; the sensitivity analysis then adds the second quadratic, over the number of drivers.

Who it is for: Risk quants and model-validation teams who want to see where amplitude estimation stands against Monte Carlo on their own risk structure; security and cryptography teams putting a date on a post-quantum migration.
Try it: Reduced versions of both run in your browser, without an account: Risk demo → Shor demo →

Add-on · Wallet exposure · Bitcoin and Ethereum

Wallet exposure: what the chain already gives away

An address goes in; an exposure tier comes out, derived from what the chain actually holds. Where the public key is on chain, a large enough quantum computer could attack it at leisure; where only its hash is, it cannot. Taproot addresses are exposed from the moment they are funded, because the address is the key. Each verdict carries its evidence, the migration steps that apply today and those that wait for post-quantum schemes, and a threat timeline built from the published resource estimates in the Intelligence corpus.

intelligence.jos-quantum.deWallet exposure · verdict

ExposedHash shieldedContract accountNot activatedUndefined typeNot determined

Address kind	What the chain holds	Never spent	Has spent
Bitcoin P2PKH, P2WPKH 1… bc1q…	a hash of the key	Hash shielded	Exposed
Bitcoin Taproot bc1p…	the tweaked output key itself	Exposed	Exposed
Ethereum account 0x…, no code	a hash of the key	Hash shielded	Exposed
Ethereum contract 0x…, has code	code and storage	Contract account	Contract account

The tier says what is exposed, not when it will be attacked: no machine able to break secp256k1 exists today. The check reads public chain data only. It never asks for a private key, a seed phrase or an extended public key, and refuses one if it is pasted.

Who it is for: Custody, treasury and compliance teams holding Bitcoin or Ethereum on their own keys or for clients, who need to say which holdings are exposed and in what order to move them. In preparation: bulk checks across an address list, and monitoring.
Read more: The Shor demo runs the resource model behind the threat timeline; our account-type analysis of Bitcoin and Ethereum sets out the reasoning behind the tiers.

Both add-ons require an annual Intelligence subscription and are priced on request. Ask from your account page or by e-mail; we add the item to your subscription and it switches on for your account. Module details on the product page →

The conventions we hold ourselves to

A paid intelligence product is only worth its provenance discipline. These are enforced in the data layer, not left to editorial care: there is no way to record a number in this dataset without its status, its basis and its date — the schema refuses it, and the service refuses to start if a citation does not resolve.

Every figure carries a status, a basis and a date. Status says whether something was demonstrated, announced, projected or estimated. Basis says who stands behind it — a journal, the vendor, a third party, or us. A vendor’s roadmap target renders as “IBM-reported target”, never as a prediction of ours.
Demonstrated and projected never look alike. Solid lines and filled markers for results; dashed lines and open markers for plans. Roadmap dates are envelopes, not forecasts — and the envelope moves when either curve moves.
Physical and logical qubits stay separate. Different fields, different chart axes. Different vendors mean measurably different things by “logical qubit”, and comparing the raw counts is the single most common way this field is misread.
Capacity and occupancy stay separate too. A 35×35 tweezer array has 1,225 sites and ran 1,180 atoms. Those are two numbers, they are both real, and most coverage quotes whichever is larger. The schema will not let them merge.
Our own arithmetic is labelled as ours. Where we back-compute a physical-qubit count from a logical target, or a cash runway from a filing, it is marked a JoS estimate wherever it appears — including inside charts, where such a figure would otherwise travel without its caveat.
Sources carry an access date. Vendors rewrite roadmap pages without notice. Noticing that is half of what this product is for, which is why a citation without a date is not accepted.
A missing figure is an open task, never a negative. “No published two-qubit fidelity” and “poor two-qubit fidelity” are different findings, and only one of them is about the vendor. Several profiles are thin because the evidence is thin; the platform says which.

Get access

Two ways in. Subscribe online and you are inside in minutes. Or tell us what you are working on and we set up a trial by hand — it decides whether the platform is any use to you, and we would rather say so than sell you a login.

Online

Subscribe

All five modules, the full change record and the briefs, and a monthly e-mail on what changed in the dataset. Annual or monthly: twelve months paid in advance, or month by month, cancellable at the end of any month. For companies and institutions; pricing for both plans is shown when you register, before any payment.

Buy See all modules →

By request

Request a trial

Trials are set up by hand, for a stated period. Tell us below what you are deciding and we reply within one business day.

To the form ↓

Name (required)

E-mail (required)

Company (required)

Role (optional)

What are you deciding? (optional, but it is the most useful thing you can tell us)

This form adds you to no mailing list, and we share nothing with third parties. Your details are stored on Google Cloud in Frankfurt; we do not store your IP address.

JavaScript is switched off. Write to contact@jos-quantum.de with the subject “Quantum Intelligence access” and we will pick it up from there.

Data & privacy

Nothing is sent until you press the button. This page sets no cookies, runs no analytics and does no tracking.
Controller. JoS QUANTUM GmbH, Platz der Einheit 2, c/o TechQuartier, 60327 Frankfurt am Main, Germany — info@jos-quantum.de. Full details in the site notice.
What we process. Your name, e-mail address and company, and — only if you fill them in — your role and your message. Alongside these we store two technical values for abuse prevention and diagnostics: the page the request came from, and your browser’s user-agent string, truncated. We do not store your IP address in our database.
Purposes and legal basis. Processing your access request, provisioning an account if we agree terms, and contacting you by e-mail about JoS QUANTUM Intelligence and related services. Legal basis is your consent, Art. 6(1)(a) GDPR — given by ticking the box above, withdrawable at any time with effect for the future.
Processors and recipients. Your details are received and stored on Google Cloud (Cloud Run and Firestore) in region europe-west3 (Frankfurt am Main), under Google’s Cloud Data Processing Addendum (Art. 28 GDPR). Google’s infrastructure logs record the requesting IP address for up to 30 days for operation and abuse prevention; we do not copy it into our database. We do not share your details with any other third party.
If you become a subscriber. The platform itself is a separate application, also hosted on Google Cloud in Frankfurt. It uses one strictly necessary session cookie holding your account’s e-mail address or username, valid for eight hours; your browser sends it to every address under jos-quantum.de — the platform’s services and this website, which is hosted on GitLab Pages — though only the platform reads it, and stores your language preference in your browser. It runs no analytics and loads nothing from third parties beyond web fonts. Your account details — name, organisation, e-mail address, subscription state and term, or for a hand-provisioned trial the username and access end date — are processed to perform the contract, Art. 6(1)(b) GDPR, and deleted when the account is closed. The platform’s own privacy notice is reachable from its sign-in screen.
Online subscriptions. Online subscriptions are paid through Stripe Payments Europe Ltd. (Dublin, Ireland), which processes payment data as an independent controller — your card and bank details never reach us. We store your name, company, e-mail address, the Stripe customer and subscription identifiers and the state of your subscription on Google Cloud Firestore in Frankfurt (europe-west3), for the duration of the contract and then for the statutory retention periods for invoices. As part of the service, subscribers receive a monthly e-mail on what changed in the dataset (Art. 6(1)(b) GDPR), with one-click unsubscribe in every message. Sign-in links are sent to the account’s e-mail address.
Retention. Access requests that do not become accounts: reviewed and deleted at the latest 12 months after our last contact with you. Requests that become accounts follow the retention above.
Your rights. Access, rectification, erasure, restriction, portability and objection (Art. 15–21 GDPR); withdrawal of consent at any time via info@jos-quantum.de; complaint to a supervisory authority — for us, the Hessian Commissioner for Data Protection and Freedom of Information.
Third-party loads on this site. Like every page here, this one loads fonts and icons from content-delivery networks (Google Fonts, cdnjs); see the site-wide privacy notice.

Before you ask

This is a curated dataset, not a live feed. Figures are checked on a monthly sweep and a deeper quarterly pass, and every one shows when it was last current. A platform that claimed real-time hardware specifications would be claiming something no vendor publishes.
Some profiles are thin. Several vendors publish a qubit count and no fidelity, or a roadmap and no characterisation. Where the evidence is thin the profile is thin and says so — filling the gap with a plausible number would defeat the purpose.
The MERIQ layer is the published one. Criterion scores, anchors, rationales, the weakest-link cap and the horizon logic are all here and all contestable against the public rubric. The calibration weights and the benchmark library that keep a 4 meaning the same thing across quarters are what an engagement adds, and are not in the platform.
The investor module is analysis, not advice. Filings and calls are quoted verbatim with their source; figures we derive are labelled as ours. Nothing in it is a recommendation to buy or sell anything.
The interface is bilingual; the analysis is in English. Navigation, labels and this page exist in German. The curated notes, rationales and briefs are written in English, and translating them would halve how often they are updated.

See when quantum computers will deliver business value, and for which applications.