What is DARPA's Quantum Benchmarking Initiative, and who just reached its final stage?

TL;DR (as of 14 October 2026). Six companies now sit in Stage C — the final, hardware-testing stage — of DARPA's Quantum Benchmarking Initiative (QBI): Atom Computing, Diraq, IBM and IonQ, announced on 7 October 2026, alongside Microsoft and PsiQuantum, who entered through the US2QC pilot (DARPA, 7-Oct-2026). QBI is the closest thing quantum computing has to an external judge: a government evaluation team tests whether any approach can reach utility-scale operation — "computational value exceeds its cost" — by 2033. Rosetta Quantum reads the announcement the way it reads every verdict: a Stage C pass is an external audit of plans and prototypes, not a measured quantum advantage, and the public count of verified advantage claims does not move.

What is QBI, and what does it actually test?

QBI is DARPA's program to determine whether any quantum computing approach will reach utility-scale operation — defined by DARPA as a machine whose "computational value exceeds its cost" — by the year 2033 (darpa.mil, 7-Oct-2026). It launched in mid-2024 as an expansion of the earlier US2QC pilot, and has evaluated more than 20 companies since (darpa.mil, 7-Oct-2026). Program manager Joe Altepeter framed the goal as working "to separate hype from reality in quantum computing" (darpa.mil, 3-Apr-2025).

The funnel has three stages, each harder than the last. Stage A was a six-month sprint in which companies handed over comprehensive technical details of their concepts; 15 companies were named on 3 April 2025, with QuEra added on 29 April 2025, Google Quantum AI on 9 September 2025 and Quandela in May 2026 — 18 named in total (darpa.mil, 3-Apr-2025, updated; darpa.mil, 7-Oct-2026). Stage B was a year-long review of each company's R&D plan, prototype roadmap, risks and manufacturing pathway — QBI managing director Micah Stoutimore describes it as "putting the performers' plans under a microscope" — with 11 companies kept on 6 November 2025 and up to $15 million of support per team, per PostQuantum's analysis of the stage (The Quantum Insider, 7-Nov-2025; PostQuantum, Nov-2025). Stage C is where the government's independent verification and validation team tests components, subsystems, algorithms and the actual hardware. In Stoutimore's words: "Stage C is where we begin moving from plans to proof."

That structure — an evaluator with no stake, paid to be skeptical, testing the vendor's machine against the vendor's own roadmap — is what Rosetta Quantum has argued the field lacks almost everywhere else (why no one maintains a neutral verification ledger).

Who reached Stage C on 7 October 2026, and who was already inside?

Four companies joined: Atom Computing (neutral atoms), Diraq (silicon CMOS spin qubits), IBM (modular superconducting processors) and IonQ (trapped ions). They join Microsoft and PsiQuantum, who advanced through the US2QC pilot — Fast Company dates their advance to February 2025 (darpa.mil, 7-Oct-2026; Fast Company, 8-Oct-2026). Six teams, four architectures restated by DARPA, two not restated. The full map, row by row:

Two readings of that table matter more than the headline. First, absence from the 7 October list is not elimination: DARPA says it expects "more companies advancing to Stage C in the near future", and the seven remaining Stage B teams — Quantinuum among them — stay in the program on the record (darpa.mil, 7-Oct-2026). Second, the funnel is real: Google Quantum AI, Rigetti and Hewlett Packard Enterprise appear on the Stage A record but not on the Stage B list, and DARPA has published no explanation either way (darpa.mil, 3-Apr-2025, updated; The Quantum Insider, 7-Nov-2025).

What does a Stage C pass prove — and what does it not prove?

It proves that an external team with hardware access judged a plan worth testing at full scale, with real money behind the test: IonQ discloses an agreement "with a potential value of up to $300 million", running through 2029, with funding beyond the initial obligation "subject to future appropriations" (IonQ release, 7-Oct-2026). DARPA's own page publishes no dollar figure — the number comes from the vendor's release, and Rosetta prints it with that label.

It does not prove a speedup. No benchmark result, no verified advantage and no utility-scale machine is part of a Stage C selection; DARPA's team was explicit a year ago that "multiple, single, or even no participants will ultimately demonstrate a path" to utility scale (The Quantum Insider, 7-Nov-2025). The one applied quantum-advantage claim still standing in the public record — 71,313 certified random bits — predates this announcement and is unaffected by it (the only claim that survives); the map of what quantum computing is used for today, with its advantage counter, does not change either (what quantum computing is used for today).

Why does an external judge matter more than another benchmark?

Because the judge has no stake in the result, and almost every number in this industry is published by someone who does. Vendor-defined metrics like Quantum Volume and Algorithmic Qubits are self-favorable by construction (Quantum Volume vs Algorithmic Qubits). Pre-registered public benchmarks like QOBLIB fix the exam before anyone sits it, but still rely on whoever runs them (what is QOBLIB). An external verification and validation team with physical access to the machines — QBI's Stage C — is the strongest rung of that authority ladder operating anywhere in quantum today, and the whole market map of who builds what feeds into it (who is ahead in quantum in 2026).

It has one structural gap, and naming it is not a criticism: QBI's test results are not public. The field will learn who advances, not what was measured. A verdict signal without the underlying data is exactly the configuration that leaves the public record without a citable measurement — the gap a neutral verification ledger exists to fill, and the gap Rosetta works in (why no one maintains a neutral verification ledger).

Does "utility-scale by 2033" move the Q-Day clock?

No. Utility scale is an economic bar — computational value exceeding cost — and says nothing by itself about the fault-tolerant scale needed to break RSA-2048 (how many qubits to break RSA-2048). Stoutimore now says "we increasingly expect that someone will build a utility-scale quantum computer by 2033", while adding "what we don't yet know is which team or teams, or which technical approach or approaches, will get there" (darpa.mil, 7-Oct-2026). Rosetta's sealed Q-Day position — floor 2031, central window 2033-2039, late bound 2040, seal RQ-PRED-QDAY-001 — will not be revised on this announcement; the quote will enter the file as one input for the committed semiannual review, around April 2027 (the Q-Day bet). Diraq's own expectation of reaching utility scale by 2031 is the company's forecast, stated in the future tense, with no public benchmark attached to it yet (Fast Company, 8-Oct-2026).

What we don't know

• Whether any Stage C system will pass. DARPA's own framing allows "even no participants" to demonstrate a path to utility scale. A filter that cannot fail everyone is not a filter; this one can.
• What the IV&V team will measure, or against what thresholds. The test criteria and results are not public, and DARPA has not committed to publishing them.
• The money, beyond vendor statements. The $300 million cap comes from IonQ's release and press coverage, not from darpa.mil; the $15 million Stage B figure comes from PostQuantum's analysis. We found no DARPA page stating either number.
• Which Stage B teams will advance next. DARPA says more will, "in the near future", without naming them.
• Why Google Quantum AI, Rigetti and HPE appear in Stage A but not on the Stage B list. DARPA has not said, and absence of an announcement is not an announcement.
• Whether Microsoft and PsiQuantum have passed any Stage C test since February 2025. No public result exists in either direction.
• Whether "computational value exceeds cost" will be measured publicly. No formula, workload set or audit trail for that definition has been published.

Rosetta Quantum publishes verdicts with raw, reproducible data. This is educational content, not a product claim. Stage assignments and quotes verified against the sources below on 14 October 2026.

Sources

• DARPA. Four more teams enter Quantum Benchmarking Initiative final stage, 7 Oct 2026 — darpa.mil
• DARPA. Companies targeting quantum computers, 3 Apr 2025 (updated 29 Apr 2025 and 9 Sep 2025) — darpa.mil
• The Quantum Insider. DARPA Advances Quantum Computing Initiative, 7 Nov 2025 — thequantuminsider.com
• IonQ. IonQ Advances to Final Stage of DARPA's Quantum Benchmarking Initiative, 7 Oct 2026 — investors.ionq.com
• Fast Company. DARPA just picked four early winners for its quantum initiative, 8 Oct 2026 — fastcompany.com
• PostQuantum. DARPA's Quantum Benchmarking Initiative (QBI): Stage B, Nov 2025 — postquantum.com
• Rosetta Quantum — why does no one maintain a neutral quantum verification ledger
• Rosetta Quantum — who is ahead in quantum computing in 2026
• Rosetta Quantum — what is quantum computing used for today
• Rosetta Quantum — the only quantum advantage claim that survives today
• Rosetta Quantum — when is Q-Day coming, based on what can be measured today
• Rosetta Quantum — Quantum Volume vs Algorithmic Qubits
• Rosetta Quantum — what is QOBLIB and how do you use a pre-registered benchmark
• Rosetta Quantum — how many qubits does it take to break RSA-2048