Connect with us

NEWS

Quantinuum Helios Record Frames a 2029 Fault-Tolerance Bet

Helios stored a logical qubit at five errors per million cycles, a 12-fold slope Quantinuum now treats as its path to a 2029 fault-tolerant machine.

Published

on

Quantinuum’s Helios computer stored an error-corrected qubit at five failures per million check cycles, the lowest logical memory error yet posted. The 98-qubit trapped-ion machine used a color code that got cleaner as it grew, and the company is now tying that slope to a full fault-tolerant system in 2029.

Each time the team thickened the code, memory errors fell about twelve-fold. That slope is the product Quantinuum is selling: a path to useful machines with tens of thousands of ions, not a claim that Helios already is one.

Five Errors in a Million Cycles

On October 19, 2026, Quantinuum said Helios had reached a logical memory fidelity of 99.9995% per error-check cycle. That is five uncorrectable errors per million rounds on a stored logical qubit. The company called it a tenfold gain on its own September mark, and more than a hundredfold better than the superconducting record it used for comparison.

The underlying paper, posted to arXiv on October 16 by a team that includes Ali Lavasani and David Hayes, reports below-threshold triangular color codes on Helios at distances 3, 5 and 7. Distance is the code’s toughness. A larger distance spends more physical qubits to protect one logical bit of information.

Helios holds 98 trapped barium ions, steered with lasers and electric fields. The distance-7 block ate 73 of them, data ions plus the extra ions used to measure errors. The machine can host one such block at a time. It cannot host two.

The test ran 20 rounds of syndrome extraction, the repeated parity checks that look for errors without reading the stored bit. The posted memory-error estimates are 5.5 x 10-4 at distance 3, 3.8 x 10-5 at distance 5, and 5 x 10-6 at distance 7, the last with a one-standard-deviation range of +4/-2 x 10-6.

THE HELIOS COLOR-CODE RESULTS

Code distance Physical qubits in the block Memory error per cycle Two-qubit Clifford error
3 13 5.5 x 10-4 7 x 10-4
5 37 3.8 x 10-5 7 x 10-5
7 73 5 x 10-6 Not run

Unencoded two-qubit circuits on the same hardware failed at 9.5 x 10-4 per gate. Encoding helped, but only once the code got large enough that the extra circuitry did not eat the gain. Distance 3 barely moved the gate number. Distance 5 brought it down to 7 x 10-5.

Error Rates Fell Twelve-Fold at Each Step

The figure Quantinuum is carrying into its next machines is not the five-in-a-million headline. It is Lambda, the factor by which logical errors drop each time the code distance steps up. For memory, the fit is 12. For the two-qubit Clifford circuits, the fit is 9, with a wide uncertainty band of +8/-3 around that 9.

Below-threshold behavior is the old theoretical promise: once physical noise is low enough, spending more qubits should make the logical bit exponentially cleaner. Google showed a version of that in late 2024 with a much smaller Lambda. Helios is the first published color-code run in which that slope is an order of magnitude per step.

Hayes, a Quantinuum physicist on the paper, said the record also comes from tighter control of the physical ions. When an ion’s energy drifts in a way that would spoil a gate, a laser pulse now knocks it back. The color code then uses correlations across the block to catch what is left. The paper groups those tricks as circuit-level and hardware-level leakage reduction.

Steven Flammia, a Virginia Tech physicist who did not work on the project, called it an impressive technical advance. Murphy Niu at the University of California, Santa Barbara, called it a major step toward computers that can keep correcting themselves as they run. Both were talking about a storage result that improves when the block grows, which is the condition any later computer has to meet.

What a Megaquop Machine Has to Do

John Preskill at Caltech, who coined NISQ in 2018, spent a December 11, 2024 keynote naming the next rung. He wanted a machine that can run about a million quantum operations, or somewhere near it, with error correction actually working. He called it a megaquop machine. Mega, he said, means in the vicinity of a million, not a precise count.

Naively, a megaquop machine would have an error rate per logical gate of order 10-6, which we don’t expect to achieve anytime soon without using error correction and fault-tolerant operation.

John Preskill, California Institute of Technology, Q2B 2024 keynote

He put a rough shape on the workload: on the order of 100 logical qubits, at a depth of about 10,000. He guessed that tens of thousands of high-quality physical qubits could suffice. Helios’s distance-7 memory rate, 5 x 10-6 per cycle, sits next to that 10-6 gate target. The memory test is not a million-gate algorithm.

PRESKILL’S MEGAQUOP BAR

  • The unit: About one million reliable quantum operations in a single run, a length of computation rather than a qubit count.
  • The error: On the order of 10-6 per logical gate, reached with error correction rather than by polishing physical devices alone.
  • The shape: Roughly 100 logical qubits at a circuit depth of about 10,000, enough, in his view, to simulate many-qubit physics that NISQ devices cannot hold.
  • The hardware guess: Tens of thousands of high-quality physical qubits, with Rydberg atoms in optical tweezers named as a likely early modality if they kept improving.

Preskill would not promise commercial jobs at that scale. He expected scientists to learn from dynamics they cannot fake classically. Helios has now posted a memory number in the 10-6 band. It has not posted 100 logical qubits, and it has not posted a depth of 10,000.

Willow’s Factor of Two Set the Last Public Bar

The superconducting number the field still quotes is Google Quantum AI’s distance-7 surface-code memory on Willow, published in Nature in December 2024. That logical memory showed 0.143% error per cycle on a 101-qubit distance-7 code, with Lambda of 2.14 as the distance stepped from 3 to 5 to 7.

A factor of two per step is below threshold. It is also a slow climb. Helios’s fitted memory Lambda of 12 is the contrast Quantinuum wants investors and customers to hold. The two experiments are not identical: different codes, different cycle times, different machines. The published per-cycle memory errors still sit two and three orders of magnitude apart.

THREE MACHINES, THREE ERROR STORIES

System Status Physical qubits Logical layout Posted or planned error
Helios color code Measured, October 2026 98 on the machine; 73 in the distance-7 block One logical qubit at distance 7 5 x 10-6 per memory cycle
Willow surface code Measured, December 2024 101 in the distance-7 code One logical qubit at distance 7 0.143% per cycle
QuEra Libra Planned for 2028 More than 10,000 256 logical qubits 10-6 logical error target

Quantinuum then ran the Lambda-12 slope out to a utility-scale error of 10-12. On that projection, a color-code ion machine would need a distance of 21, or about 30,000 physical qubits, to field 100 logical qubits. The superconducting color-code result it compared against would need a distance of 109, or about 900,000 physical qubits, a gap the company puts at roughly 30 times. That 10-12 target is far stricter than Preskill’s 10-6 megaquop gate, and far stricter than Libra’s planned 10-6. It is a resource argument, not a machine that exists.

The Gate Circuits Did Not Fit

Two distance-7 color-code blocks would need 73 plus 73 physical qubits, which is 146. Helios has 98. The paper did not run the distance-7 gate test because the circuits do not fit. The two-qubit result that does exist, 7 x 10-5 at distance 5, is about fourteen times the distance-7 memory rate. Storage and logic are different jobs on this hardware.

That gap is the objection that survives the press release. A logical bit that sits still through 20 check rounds is not a computer. Leaving the noisy era means the same slope has to hold when those bits are asked to interact, and when the decoder has to answer in time to choose the next gate.

We felt a mix of excitement and suspense every time the results came in. When errors are this rare, each one really matters.

Adam Reed, Lead Physicist, Commercial Operations, Quantinuum blog, October 19, 2026

The company says two of the three pieces of a Clifford-plus-T stack now scale below threshold on Helios: the check-cycle gadgets and the two-qubit Clifford gates. Magic-state preparation, the extra resource that makes the gate set universal, still lacks a below-threshold scaling proof on this code. Quantinuum points to a 2025 color-code magic-state paper as the start of that work, not the finish. Google, in the same month as the Helios post, reported a 40-fold cut in cultivated T-state errors on Willow, a different bottleneck on a different machine.

WHAT HELIOS STILL HAS NOT RUN

  • Distance-7 gates: Two encoded qubits at the record code size, and a logical operation between them, need more ions than Helios holds.
  • Magic-state slope: No group has yet shown below-threshold scaling for color-code magic-state preparation, the missing universal-gate ingredient on this stack.
  • A hundred logical qubits: Thomas Wong at Creighton University noted that known megaquop applications want on the order of 100 logical qubits, while Helios has on the order of 100 physical qubits.
  • Deep circuits: Flammia said the memory rates fit a megaquop error band, and that the open question is whether they stay that low once the logical qubits run harder math.

Hayes called the Helios work homework for the next boxes. “We are doing all of our homework up front,” he said. “It’s super exciting.”

QuEra Booked 2028 on Neutral Atoms

Preskill’s 2024 guess for a fast megaquop path was Rydberg atoms in optical tweezers, if they kept gaining both scale and quality. QuEra Computing, whose devices trap cold atoms, has since put a date on that idea. Its public roadmap sets Libra, a megaquop-class machine, for 2028, with 256 logical qubits on more than 10,000 atoms and a 10-6 logical error target. A later gigaquop system is listed for 2028/29 at 1,000-plus logical qubits and more than 20,000 physical qubits.

That is a different bet. QuEra is promising many logical qubits at the same 10-6 error Helios has now shown in memory, on a cloud service through Amazon Braket, in 2028. Quantinuum is promising a steeper error slope on far fewer physical qubits, with Apollo, its first universal fault-tolerant computer, planned for 2029. Helios, in the company’s phrasing, is a pathfinder for that date.

Sol, the system after Helios, is due the next year. Hayes said it will be nearly twice as large. The Quantinuum blog says Sol moves the trap from Helios’s layout to a grid, shortening how far ions have to travel and keeping all-to-all connections as the chip grows. Until Sol runs, the grid is a working device in prototype footage, not the computer customers will book.

THE ROAD TO A MEGAQUOP DATE

  1. December 11, 2024: Preskill names the megaquop machine at Q2B and sets a rough bar of 10-6 per logical gate.
  2. December 2024: Google posts Willow’s distance-7 surface-code memory at 0.143% error per cycle and Lambda 2.14.
  3. November 5, 2025: Quantinuum launches Helios as a 98-qubit commercial trapped-ion machine.
  4. September 2026: Helios sets a prior logical record with the Helix code, the mark the October result beats tenfold on memory.
  5. October 16, 2026: The color-code paper goes to arXiv; three days later Quantinuum posts the 99.9995% memory figure and the Lambda-12 slope.
  6. 2027: Sol is due, nearly twice Helios in size, on a 2D grid trap.
  7. 2028: QuEra plans Libra as a 256-logical-qubit megaquop system.
  8. 2029: Quantinuum plans Apollo as its first universal fault-tolerant computer.

Wong’s arithmetic still sits on the table. Helios proved that one logical ion-block can be held at five errors per million cycles, and that making the block bigger made it cleaner by a factor of 12. It did not prove that 100 of those blocks can talk to each other for 10,000 steps. Sol is the first Quantinuum box sized to try the gate test Helios could not fit. Apollo is the date the company has now attached to the slope.

Frequently Asked Questions

What Is a Logical Qubit?

A logical qubit is a protected bit of quantum information encoded across many physical qubits so that some errors can be found and fixed. In the Helios color-code tests, code distance is the smallest number of physical errors that can flip that logical bit, which means a distance-7 block can correct up to three physical errors in the right pattern. The extra ions in each block are measurement qubits used to extract an error syndrome without reading the stored state.

What Is a Megaquop Machine?

A megaquop machine is Preskill’s name for an early error-corrected computer that can finish about a million quantum operations in one run. He introduced it in the Q2B 2024 keynote and in a 2025 paper, and he was explicit that mega means somewhere near a million rather than exactly 1,000,000. He also said he could not promise commercially viable applications at that scale, only that scientists should be able to simulate many-qubit dynamics beyond NISQ devices.

How Many Physical Qubits Does Helios Have?

Helios is a 98-qubit trapped-ion processor that Quantinuum launched on November 5, 2025, using barium-137 ions in a QCCD trap with laser and electric-field control. The October 2026 color-code memory block at distance 7 used 73 of those ions, so the record logical qubit occupied most of the machine. Two such blocks cannot be loaded at once on the current chip.

What Error Rate Did Helios’s Color Code Reach?

The distance-7 memory error is 5 x 10-6 per syndrome-extraction cycle, reported with a one-standard-deviation Wilson interval of +4/-2 x 10-6, after 20 rounds per shot. The matching two-qubit Clifford error at distance 5 is 7 x 10-5 per logical gate, and the distance-7 version of that gate test was not run. Those two numbers are the memory record and the best encoded gate result, not a single combined rate.

Harry is the editor and lead writer of STUDIO ONE NETWORKS, an independent title he owns and runs himself. Ten years in journalism, reporting first and editing later, taught him that entertainment and business are one beat seen from two sides: a box office figure is a company number, a streaming deal is a contract, a casting rumour is not a story until someone puts their name to it. He works from the record, whether that is a distributor's statement, a licensing agreement, an interview transcript or a set of published ratings, and checks every number against it before publication. The same rule holds for the rest of the site, which covers news, technology, science, sports, lifestyle, travel, auto and gaming for an audience spread across the world. When he gets something wrong, the article is corrected and the change is noted and dated, under a corrections policy anyone can read. Reader mail is answered by him at support@studioonenetworks.com.

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Trending