Three years after the start of the Quantum Benchmarking program, roughly ten of nearly 200 initial candidate application instances had been carried through to end-to-end resource estimates. Yudong Cao, CTO partner at Zapata, offers a candid critique of the program, stating, “We knew that we could still go further, but the program has to end and we have to turn in our deliverable.” This narrowing reflects the intense labor required to move from promising concepts to concrete resource estimates for even simplified problems.
DARPA Quantum Benchmarking Initiative: Initial Scope & Timeline
Zapata, alongside L3Harris and other organizations, initially identified between 150 and 200 candidate application instances as potentially benefiting from quantum computing, forming the initial evaluation funnel for the Quantum Benchmarking program. Detailed interviews with domain experts across seven to nine fields revealed chemistry and differential equations as particularly promising avenues for further investigation. The program’s focus wasn’t simply on identifying if quantum computers could solve problems, but which problems warranted the substantial investment required to develop the necessary hardware and software.
Cao describes the resulting map as a snapshot taken in March 2025, representing analysis that was not fully optimized. The team encountered a challenge by applying a computationally expensive method to individual instances before determining their overall worth.
Cao advocates for a shift towards a lightweight predictive model for quantum amenability, trained on the limited number of precise data points generated by programs like the Quantum Benchmarking Initiative. He believes this model could then be used to generalize across a wider range of potential applications, streamlining the discovery process and focusing resources on the most promising areas.
The program’s timeline ultimately forced a premature conclusion; “There are orders of magnitude improvement you can make if you just give us six more months. But we had to stop because that was when the buzzer rang.” This highlights the tension between the desire for thoroughness and the practical constraints of a fixed-duration research project, leaving some potential gains unrealized.
We knew that we could still go further, but the program has to end and we have to turn in our deliverable.
Yudong Cao, CTO at Zapata
L3Harris & Zapata Identify Narrowed Application Focus
Initial assessments of potential quantum applications yielded a surprisingly large number of candidate instances, with Zapata, L3Harris, and collaborating organizations identifying between 150 and 200 possibilities as the starting point for detailed evaluation. This represented the top of the analytical “funnel,” a broad sweep intended to capture any area where quantum algorithms might offer an advantage, but the sheer volume quickly demanded prioritization. The three-year effort ultimately delivered a narrowed set of ten specific applications, each subjected to rigorous end-to-end resource estimation.
Cao argues that the program’s structure forced a premature conclusion, despite the potential for significant gains with additional time. He believes such a model could streamline the evaluation process and provide a more accurate forecast of future possibilities, but it would require a different approach than the one employed during the benchmarking initiative.
Resource Estimation Challenges & The “False Precision” Problem
Following initial interviews with experts, chemistry and differential equations emerged as the two fields most promising for further investigation. Roughly ten of those initial candidates underwent complete resource estimation, detailing physical qubit requirements and projected runtime, yet Cao argues the resulting data presents a misleading sense of accuracy. A reference to the program’s visual representation of application resource needs is that each data point represents a calculation halted by a fixed deadline.
The core of the challenge, Cao explains, lies in translating technical specifications into terms meaningful to decision-makers outside the quantum field. “If you sit across the table from someone who runs logistics for a living, they don’t know what it means for an application to take 5 million qubits and five hours,” he said, highlighting the disconnect between quantum metrics and practical business considerations. “They speak in a very different language.” This disconnect creates a “false precision” problem, where detailed resource estimates are presented without context, hindering effective evaluation of quantum computing’s real-world applicability.
If you sit across the table from someone who runs logistics for a living, they don’t know what it means for an application to take 5 million qubits and five hours.
Yudong Cao, CTO at Zapata
Predictive Modeling for Quantum Amenability & Scalable Discovery
The three-year Quantum Benchmarking program, he argues, prioritized completing resource estimation on a limited set of roughly ten candidate applications rather than optimizing the evaluation process itself. Cao presented this assessment during his talk, “Quantum Applications, Algorithms, Circuits,” at the Quantum Alliance 2026 Annual Meeting. The resulting plot of application instances, while visually tidy, represents a snapshot taken in March 2025, before full optimization could occur.
He explains that the work stopped because the program ended, not because the analysis was finished. He believes the current method applies an expensive process to individual instances, consuming years to determine their worth, and this approach is the reverse of what is needed to efficiently discover viable quantum use cases.
