Innovation & Tech

Moth's Tessa Stores 1,568 Values Across 120 Qubits on IBM Chip

Moth's Tessa encoded and recovered 1,568 values on IBM's 120-qubit Phoenix chip without SWAP gates, hitting 0.96 correlation in single-basis tests.

By Grace Kim

4 min read

Moth’s Tessa Encodes and Recovers 1,568 Values on a 120-Qubit Chip
Moth’s Tessa Encodes and Recovers 1,568 Values on a 120-Qubit ChipTheDigitalArtist / pixabay

What's News

  • Moth announced Tessa on November 16, 2026, encoding and recovering 1,568 values on IBM's 120-qubit Phoenix Nighthawk r2.
  • Single-basis runs reached 0.96 value correlation on Phoenix with mean absolute error of 0.130, using 8,192 shots per basis and no error mitigation.
  • The pair-mode run scored 0.69 value correlation and used 1,456 two-qubit CZ gates, double the single-basis count.
  • Phoenix recovered the 784-nucleotide MT-CO3 sequence with 92.1% of nucleotides exactly correct at 32,768 shots per measurement.
  • Principal hardware experiments date to September 26, 2026, per the repository's run log.

Moth says its Tessa method encoded and recovered 1,568 real values across all 120 qubits of IBM's Phoenix Nighthawk r2 processor — a claimed record for loading arbitrary classical data onto a quantum chip and getting it back. The company announced the result on November 16, 2026, targeting one of quantum machine learning's stubborn bottlenecks: getting classical data into fragile qubits on sparsely connected hardware.

Daniel Bultrini's paper scopes the claim precisely. It counts only experiments that loaded freely chosen numeric lists onto physical processors and retrieved the individual values — not runs that returned a mere derived quantity. Under that definition, QCrank previously recovered 128 values from 12 trapped-ion qubits. A separate Fez experiment recovered 1,024 pixels from 13 qubits, but it leaned on a trained tensor-network approximation rather than an arbitrary list of real values, so it doesn't compare directly. The paper calls Tessa's run the largest reported under its definition.

How Does Tessa Use the Chip's Wiring?

Tessa — short for Tessellated QPIXL — alternates data and address qubits in a checkerboard pattern laid over the processor's physical connections. Each data qubit is controlled only by neighboring address qubits, which nearby data qubits can share. Every additional address neighbor doubles a data qubit's storage, so an interior qubit with four neighbors holds 16 values.

The construction runs on any bipartite coupling graph, where every coupler joins a data qubit to an address qubit. Because it uses the device's own layout, Tessa needs no SWAP gates, no routing and no ancilla, or helper, qubits. The paper proves that values on each data qubit stay independently recoverable despite the shared controls.

All values stored in a single basis decode from one round of computational-basis measurements across the chip. Tessa can also store a second number in each slot's phase. The Phoenix run used this pair mode for an image's brightness value and hue — two numbers in each of 784 used slots, with zeros filling spare slots so every qubit participated.

That density carries costs. Pair mode adds measurement bases and doubles circuit depth. More broadly, shots get divided among local address configurations, so the number needed for equivalent accuracy grows exponentially with local coupling degree — though the experiments kept that degree small. Density, in short, is not free.

What Did the Hardware Actually Recover?

Single-basis tests used 8,192 shots per basis to recover a 28-by-28 QCAT image on 784 of Phoenix's 792 slots and a 21-by-21 image on 441 of Fez's 448 slots. Raw hardware value correlations hit 0.96 on Phoenix and 0.95 on Fez, with mean absolute errors of 0.130 and 0.131 respectively. Dynamic-range recovery — which uses summary statistics of the input rather than the input itself — cut those errors to 0.075 and 0.078. The experiments used no error mitigation.

The 1,568-value pair run was less accurate: a value correlation of 0.69, a value mean absolute error of 0.152 and a hue mean absolute error of 0.092 on Phoenix. Its circuit needed 1,456 two-qubit CZ gates, versus 728 for the single-basis Phoenix run. Full-chip retrieval worked on both coupling graphs, but the second value in each slot came with higher circuit demands and lower reconstruction quality.

What About Biological Data?

Beyond pixel art, the experiments covered audio, MIDI, DNA and protein amino-acid sequences, including p53. The biological sequences were picked for their lengths or names, not their biology — these were data-handling tests, not biological analyses.

At 32,768 shots per measurement, Phoenix recovered the 784-nucleotide MT-CO3 sequence with 92.1% of nucleotides exactly correct without post-processing, and 92.3% with the best tested dynamic-range recovery, selected using the true sequence. Densely packed protein sequences recovered worse. The paper cautions that even one incorrect nucleotide or amino acid can have serious biological consequences.

Moth says Tessa is available for image-processing experiments on its Atlas platform. The repository's documentation provides the library, raw hardware counts, job records, simulation outputs and notebooks that regenerate the paper's figures and tables without an IBM account. The run log dates the main Phoenix and Fez QCAT experiments to September 26, 2026, with later genetic and pixel-art runs in October 2026 — separating the hardware work from the November announcement.

Tessa's immediate contribution is a hardware-aware way to load and verify classical data across an entire processor. For quantum machine learning, the open question is whether later computations can exploit the encoded state without paying for full reconstruction — while keeping shot growth and the errors of denser pair encoding in check.

Source: unite.ai

Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.

2,174 articles

Related articles

« Previous articleNext article »