Quantum supremacy in 2026 — a reality check from someone who ran real circuits

I've spent real time with two of the most advanced quantum computers that are actually reachable over the internet: IBM's ibm_fez (156 qubits) and Origin Quantum's WK_C180 (180 qubits). We used both to generate the artwork for Quantum Genesis — running real circuits, collecting real measurement data, and turning quantum noise into abstract art.

That hands-on time gives a different view of the field than most press releases do. This is that view: what "quantum supremacy" actually claims, where the hardware genuinely stands in 2026, what a developer can actually run today, and what I found running identical circuits on both platforms.

What "quantum supremacy" really claims

The term, coined by John Preskill in 2012, is narrow and specific. It means a quantum computer has performed a computation that no classical computer can do in a reasonable time. Not "could theoretically", but actually did it, with verifiable results. It doesn't mean quantum computers beat classical ones at everything — just that there's at least one task where they win. The task doesn't even have to be useful; it just has to be genuinely hard classically.

That distinction is the whole story of this field so far. Supremacy has been demonstrated for artificial benchmarks. Nobody has demonstrated a quantum advantage on a practical problem that anyone outside a physics lab would care about.

Google's Sycamore, and after

In October 2019, Google published a Nature paper claiming supremacy with its 53-qubit Sycamore processor: sampling task done in 200 seconds that they said would take a supercomputer 10,000 years. IBM immediately pushed back, arguing their Summit machine could do it in about 2.5 days with clever classical algorithms and enough disk. The argument ran for years while classical simulation kept improving.

In 2024 Google came back with Willow (105 qubits), doing a random circuit sampling task they estimated at 10^25 years classically — a number larger than the age of the universe. More important than the headline number: Willow showed that adding qubits reduced the logical error rate via surface codes, which crosses a genuinely critical threshold for quantum error correction.

The honest summary: supremacy has been demonstrated for sampling tasks with no known practical application. Its significance is foundational — proof that quantum mechanics offers a computational resource classical physics can't efficiently simulate. That's science, not a product.

IBM's roadmap: the pivot from Eagle to Heron

IBM has been the most transparent about its hardware path:

Processor	Qubits	Year	Key achievement
Eagle	127	2021	First processor over 100 qubits
Osprey	433	2022	3x qubit count increase
Condor	1,121	2023	First processor over 1,000 qubits
Heron	133-156	2024	Focus on quality over quantity
Starling	TBD	2025-2026	Error-corrected logical qubits

Watch the pivot at Heron. After hitting 1,121 qubits with Condor, IBM reversed direction: the Heron processors (ibm_fez, ibm_torino, ibm_marrakesh, ibm_kingston) have fewer qubits but materially better gate fidelities and connectivity. IBM concluded that 1,000 noisy qubits are worth less than 156 clean ones.

That's where the whole industry is heading: quality over quantity. A qubit that errors 1% of the time is far more useful than one erroring 5%, because error-correction overhead scales exponentially with the physical error rate.

Origin Quantum's WK_C180

Origin Quantum (Hefei, China) gets far less Western press, but the hardware is real. Their WK_C180 — which we believe belongs to the Wukong / "Monkey King" processor family — is a 180-qubit superconducting chip reachable through their cloud, via the pyqpanda3 SDK.

Our experience:

• Access: a free tier exists with an API key, and registration is straightforward.
• SDK: pyqpanda3 is Python-native and decently documented (mostly in Chinese, some English).
• Queue times: wildly variable — sometimes seconds, sometimes hours. Less predictable than IBM.
• Results: the measurement distributions had a different noise character than IBM's — more entropy in some circuit configurations, which made for visually richer art.
• Gotcha: the measure() syntax requires explicit classical bit specification — measure([0,1], [0,1]) — which tripped us up at first.

Origin also offers the PQPUMESH8 chip and several simulators (full_amplitude, partial_amplitude, single_amplitude). The real chips are the interesting ones: a simulator gives you the same distributions as a classical RNG. NFTs #1-18 ran on WK_C180, and you can see the different aesthetic in those early pieces versus the IBM-generated #19-100.

What you can and can't run today

Things that work:

• Random circuit sampling — the supremacy benchmark. Quantum machines genuinely win here, though it has no direct application.
• True random number generation — quantum measurements are fundamentally random. This is what we built Quantum Genesis on, and it's commercially viable today.
• Small quantum chemistry — molecules up to ~20-30 orbitals via variational methods (VQE), sometimes matching or beating classical methods, though only at sizes where classical methods also work.
• QAOA on small optimization problems — under about 30 variables it works; beyond that classical solvers still win.
• Educational and research circuits — understanding the physics, testing error correction, calibrating hardware. This is the real workhorse use of most access today.

Things that don't work yet:

• Breaking encryption — Shor's algorithm needs thousands of error-corrected logical qubits. We have zero general-purpose error-corrected logical qubits.
• Drug discovery at scale — the molecules that matter (proteins, complex catalysts) need far more qubits and fidelity than exist.
• Machine learning speedup — for all the papers, no quantum ML algorithm has shown practical advantage on real hardware.
• Financial optimization — same story; theoretical speedups don't survive contact with current hardware.

Running identical circuits on both platforms

For Quantum Genesis we ran the same circuit structure on IBM and Origin hardware. Our circuits used Hadamard gates (superposition) and CNOT gates (entanglement) over 4-6 qubits, measured with 4096 shots per NFT:

# Simplified circuit for NFT seed generation
# 1. Put all qubits in superposition (H gates)
# 2. Create entanglement (CNOT ladder)
# 3. Add rotation gates parameterized by token ID
# 4. Measure all qubits, 4096 shots
# 5. Hash the measurement distribution → seed

IBM (ibm_fez, ibm_torino):

• Queue time: 1-5 minutes, generally predictable
• Execution: ~19 seconds per circuit on ibm_fez
• Results: consistent shot-to-shot; noise signature stable within a ~24h calibration cycle
• SDK (Qiskit): mature, well-documented, excellent tooling
• Transpilation required — IBM chips have limited connectivity, so logical circuits get mapped to physical topology

Origin Quantum (WK_C180):

• Queue time: highly variable (seconds to hours)
• Execution: comparable once it ran
• Results: different noise profile — higher entropy in some configurations, which we attribute to different decoherence characteristics
• SDK (pyqpanda3): functional but less mature than Qiskit; docs mostly Chinese

The art difference was tangible: Origin's measurements produced seeds with more entropy variation, giving more varied palettes and shape distributions. IBM's were more consistent but still genuinely random.

Where that leaves the hype

Let me be direct about 2026.

The hype: "quantum computers will break all encryption" — not for at least a decade, likely longer. "Supremacy means quantum computers are faster" — only on artificial benchmarks. "1000+ qubits means we're almost there" — qubit count is meaningless without quality metrics. "Quantum AI will revolutionize ML" — no evidence of that on real hardware.

The reality: we're solidly in the NISQ (Noisy Intermediate-Scale Quantum) era and will be for years. Error rates are coming down but remain orders of magnitude too high for fault-tolerant computation. The honest near-term applications are quantum random number generation, small chemistry, and fundamental research. IBM's pivot from Condor to Heron is the clearest signal of that truth — quality over quantity.

But I'm genuinely optimistic, and not for the usual reasons:

• Error correction is working. Willow crossed the threshold where more qubits means fewer errors. Everything else builds on that.
• The hardware is accessible. We ran circuits on processors in New York and Hefei from a laptop in Brazil. The democratization is remarkable.
• The pipeline is maturing. Five years ago this was a niche academic field; now there are SDKs, cloud platforms, courses, and communities.
• Quantum randomness has immediate, real value. Quantum Genesis is a working counterexample to "you can't build anything with quantum hardware today" — you just have to be creative about what you build.

That's what the collection is really for me: a timestamp of the era. Each piece is a permanent record of what these processors could do in early 2026 — noise, limitations, and all. When fault-tolerant machines arrive years from now, these hundred artworks will read as artifacts of the pioneer era. I wrote up how those measurements become art in the generator post — and if you've run real circuits yourself and seen different numbers, I'd genuinely like to compare notes.