Pith. sign in

REVIEW 3 major objections 7 minor 50 references

Quantum Circuit Vision: Cost-Aware Evaluation of Visual AI Agents for Quantum Code Generation

T0 review · 3 major / 7 minor · reviewed 2026-07-14 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A mid-tier multimodal model turns quantum circuit diagrams into unitarily verified code at about 91% accuracy on a core set for roughly one-fifth the cost of the strongest model, while routing models beats chain-of-thought as a cost lever.

desk verdict Solid Claude-family + Qiskit-drawer evidence for diagram-to-verified-code, with a useful dataset and honest stats; the cascade slogan is the softest piece. read the letter →

arxiv 2607.10057 v1 pith:AZW23J5Z submitted 2026-07-11 quant-ph cs.AIcs.CVcs.LG

Dongping Liu , Aoyu Zhang , Luyao Zhang This is my paper
classification quant-phcs.AIcs.CVcs.LG
keywords QuantumComputingMultimodalLLMsVisualCodeGenerationCost-AwareEvaluationBenchmarkUnitaryfidelityCascaderoutingcircuitdiagrams
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether multimodal AI can read quantum circuit diagrams and emit executable code that is objectively correct—and what that costs. It introduces a 132-circuit benchmark spanning thirteen categories with Amazon Braket ground truth and unitary-fidelity verification, then evaluates three Claude-family capability-cost tiers with five repeated trials on a core subset. The mid-tier model (Sonnet 4.6) reaches about 91% pass rate under simple vision prompting at roughly 18% of the top model’s per-call cost, and the top model’s accuracy edge is not statistically significant. Circuit depth, not qubit count, predicts failure; chain-of-thought prompting does not help in aggregate. A cheap-to-expensive cascade reaches 84% accuracy at 38% of single-model cost. The work releases the dataset, cost logs, and verification scripts so diagram-to-code translation for quantum software can be measured and automated rather than done by hand.

What carries the argument

QCV-Dataset paired with a three-level oracle: syntax check, sandboxed Braket Circuit execution, and unitary fidelity F=|Tr(U†_gt U_gen)|/2^n ≥0.99 from full basis-state simulation. This objective pass/fail—not an LLM judge—grounds the cost–accuracy frontier and the cascade routing claim.

What would settle it

Re-run the 21-circuit core with n≥5 trials and the cascade protocol on at least two non-Claude image-capable models, and redraw the same circuits in TikZ or hand-drawn style; if mid-tier cost–accuracy dominance or the null CoT result disappears under those conditions, the central deployment claim fails.

Watch

Extended reading notes

Core claim

On the 21-circuit core benchmark with n=5 trials, Sonnet 4.6 under Basic Vision achieves 91.4%±5.2% pass rate (unitary fidelity ≥0.99) at 1.30× credits per call—about 18% of Opus 4.6’s per-call cost—while Opus’s accuracy advantage is not statistically significant (paired t, p=0.083). Across 132 circuits, strong models reach roughly 77–78%. Logistic regression finds circuit depth, not qubit count, is the primary structural predictor of failure (depth p<0.001; qubit count p=0.20 jointly). Chain-of-thought has no significant effect (all p>0.18). A Haiku→Sonnet→Opus cascade yields 84% accuracy at 38% of Opus-only cost, so model routing dominates prompt engineering as a cost lever.

Load-bearing premise

The cost–accuracy and “routing beats prompting” conclusions rest on three models from one vendor family, one CLI gateway, and diagrams from a single programmatic renderer, so they may not hold for other multimodal systems or messier real-world circuit figures.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The paper introduces Quantum Circuit Vision (QCV), a cost-aware evaluation framework and 132-circuit multimodal benchmark for testing whether frontier multimodal models can read quantum circuit diagrams and emit verified executable Amazon Braket code. Ground truth is objective unitary fidelity (F≥0.99) via a local simulator, with a three-stage syntax/execution/fidelity pipeline. On a 21-circuit core subset with n=5 trials, Claude Sonnet 4.6 under Basic Vision reaches 91.4%±5.2% pass rate at ~18% of Opus 4.6 per-call cost, with Opus’s accuracy edge not significant (paired t p=0.083). Logistic regression finds circuit depth (not qubit count) as the main structural failure predictor; chain-of-thought (TV) has no significant effect. A post-hoc Haiku→Sonnet→Opus cascade on the full set is reported at 84% accuracy and 38% of Opus-only cost. The authors release QCV-Dataset, metadata, and evaluation code.

Significance. If the core results hold, this is a useful first infrastructure paper for multimodal AI on quantum diagrams: an open 132-circuit dataset with Braket/Qiskit ground truth, unitary-oracle verification, cost/latency logging, n=5 significance testing on a core subset, and a failure taxonomy. Strengths include an external non-LLM judge (unitary fidelity), sensitivity checks on the F threshold, TOST equivalence for CoT on stronger models, explicit collinearity note (depth vs gate count r=0.862), and full reproducibility assets (HF Hub, Croissant/RAI, GitHub). The cost–accuracy framing and cascade template are timely for agentic scientific tooling, even if currently scoped to one model family and one renderer.

major comments (3)
  1. Abstract and §5.5 present the Haiku→Sonnet→Opus cascade as achieving 84% accuracy at 38% of single-model cost and as evidence that “model routing dominates prompt engineering.” That cascade is constructed post-hoc from the same single-run (n=1) 132-circuit BV outcomes used for Table 4 (57 Haiku passes; 45/75 Sonnet rescues; 10/30 Opus rescues → 112/132), with no held-out routing policy, cross-validation of rescue rates, or multi-seed cascade. The manuscript already labels full-set/cascade numbers as “indicative” (§5.2). Please demote the cascade from a primary abstract claim to an illustrative cost-engineering example, report uncertainty (e.g., bootstrap over circuits or multi-seed re-runs on a subset), and avoid the slogan that routing “dominates” prompt engineering unless supported by a prospective comparison with matched statistical power to the CoT tests.
  2. §5.1–5.2 and Limitations §6: all primary cost–accuracy and cascade conclusions rest on three Claude-family tiers via one CLI gateway and diagrams from a single Qiskit matplotlib drawer. The paper correctly notes other gateway models lack image input and that real literature uses TikZ/Quirk/hand-drawn layouts. Because the abstract and contributions frame results as about “visual AI agents” and the cost–accuracy frontier generally, either (i) add at least one additional multimodal vendor (or a clear negative result if the gateway cannot), or (ii) systematically rephrase claims as within-family tier effects under a fixed renderer, with generalization listed as open. Visual Variants (category H) is helpful but does not substitute for cross-renderer or cross-vendor evidence.
  3. §5.2 logistic regression: depth alone explains 13.2% of deviance (p<0.001); in the joint model depth remains significant and qubit count does not (p=0.20). The text already softens “primary predictor” and notes collinearity with gate count (r=0.862). Please report full model diagnostics in the main text or appendix (coefficients with SEs/CIs, N, majority-pass definition ≥3/6, and whether category fixed effects or algorithm-family indicators change the depth result). Without that, the structural-complexity claim is directionally supported but easy to over-read relative to the modest absolute explanatory power.
minor comments (7)
  1. Table 3 vs abstract: abstract says “91% pass rate”; Table 3 reports 91.4%±5.2% under BV for Sonnet. Keep one consistent rounded figure and always attach the ±std and n=5 when quoting the core result.
  2. Table 5 / cost units: credits are platform billing units with multipliers (0.40× / 1.30× / 2.20×). State explicitly whether these are relative list prices at a fixed date and that absolute USD may change; this affects longevity of the MC and cascade percentages.
  3. §5.5 ROI vs human translation uses informal $50–200/circuit assumptions. Label as order-of-magnitude illustration, not an empirical labor study.
  4. Prompts are in Chinese (§4.1) with a brief note that EN/CN preliminary tests showed no meaningful difference. A short appendix table of EN vs CN on a few circuits would strengthen that claim.
  5. Figure 1 / dataset card: “268 annotated failure cases” in the body vs “failures (27 annotated failure cases)” in the HF configuration description (§3.1). Reconcile counts and what “27” vs “268” refer to.
  6. Eq. (3) writes F = |Tr(U†_gt U_gen)| / d with d=2^n; earlier abstract/text use the same form. Confirm global-phase tolerance is intended and that qubit ordering/endianness is fixed between GT and generated circuits in the verification scripts.
  7. Typos/consistency: “Cost-A ware” spacing in Table 1 header area; arXiv id and some related-work years (e.g., 2026 workshop/preprints) should be double-checked for final camera-ready.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: accuracy is fixed by an external unitary-fidelity oracle, costs are vendor billing units, and cascade figures are post-hoc arithmetic on observed pass/fail—not forced by definition or self-citation.

full rationale

The paper’s load-bearing claims are empirical evaluation results, not first-principles derivations that could collapse into their inputs. Pass/fail is defined by unitary fidelity F = |Tr(U†_gt U_gen)| / 2^n ≥ 0.99 against ground-truth circuits on Amazon Braket’s LocalSimulator—an external, non-LLM oracle independent of model prose or fitted accuracy targets (Eq. 3–4; §4.2). Cost metrics use platform credit tiers (0.40× / 1.30× / 2.20×) and measured latency, not parameters tuned to produce a desired frontier. Logistic regression of depth vs. qubit count and paired t-tests for CoT (n=5 core) are standard statistics on those oracle outcomes. The Haiku→Sonnet→Opus cascade (84% at 38% of Opus-only cost) is constructed from the same observed single-run pass/fail table (“Using our actual experimental data,” §5.5); that is post-hoc accounting, not a fitted input renamed as prediction, and the paper itself labels full-set/cascade numbers as indicative (n=1). Self-citations (e.g., the authors’ blockchain companion [23]) support only the domain case study, not the verification pipeline, cost model, or statistical claims. No uniqueness theorem, ansatz, or self-definitional identity underwrites the central results. The evaluation is therefore self-contained against external benchmarks; methodological caveats (single vendor family, single renderer, post-hoc cascade) are limitations of scope, not circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 2 invented entities

The work is empirical evaluation, not a first-principles derivation. Load-bearing choices are the correctness oracle (unitary fidelity threshold), the target SDK and diagram generator, the model sample, and the cascade construction from observed failures. No new physical entities are postulated; free parameters are evaluation design knobs with partial sensitivity analysis.

free parameters (4)
  • Unitary fidelity pass threshold F≥0.99 = 0.99
    Defines pass/fail for all accuracy claims. Authors report identical rates at 0.95 and a bimodal fidelity distribution, so the choice is less arbitrary than a continuous fit, but it remains a design threshold.
  • Vendor credit multipliers (Haiku 0.40×, Sonnet 1.30×, Opus 2.20×) = 0.40× / 1.30× / 2.20× per call
    All cost-efficiency and cascade percentages are in platform billing units that depend on vendor pricing at evaluation time.
  • Core subset size and composition (21 circuits, Basic/Intermediate/Advanced) = 21 circuits
    Significance tests (n=5, paired t, TOST) are confined to this hand-selected curriculum subset; full 132-circuit claims are n=1.
  • Human labor cost assumptions for ROI ($50–200/circuit, 1–4 hours) = $50–200 per circuit (estimate)
    Used only for the hybrid-workflow ROI narrative; not used in model pass-rate claims, but shapes the “>100× cheaper than humans” implication.
assumptions (5)
  • domain assumption Two circuits are correct matches iff their unitaries agree up to global phase with F=|Tr(U†_gt U_gen)|/2^n ≥ threshold.
    Standard quantum circuit equivalence practice (Burgholzer & Wille; Viamontes et al.); anchors all pass rates (§2.4, §4.2).
  • domain assumption Qiskit matplotlib-rendered PNGs are adequate visual inputs for evaluating diagram comprehension.
    All 132 images are programmatically generated this way; Visual Variants test layout but not renderer diversity (§3, Limitations).
  • domain assumption Amazon Braket SDK code generation is a representative executable target for the visual-to-code task.
    Prompts require Braket Circuit API; Qiskit GT exists but end-to-end verification wrapper is Braket-only in this work (§4.1–4.2).
  • ad hoc to paper Claude-family capability–cost tiers are a sufficient sample to discuss cost–accuracy frontiers and cascade design for frontier multimodal agents.
    Explicit experimental design choice to hold vendor fixed (§5.1, Limitations); generalization is assumed when stating routing lessons for “agentic AI systems.”
  • ad hoc to paper Default temperature, fixed BV/TV Chinese prompts, and no extra system prompts isolate model tier and CoT effects.
    Controls for prompt informativeness vs CoT (§4.1); preliminary EN/CN parity is asserted without full reported stats.
invented entities (2)
  • QCV-Dataset / Quantum Circuit Vision evaluation framework independent evidence
    purpose: Provide multimodal circuits, unitary-verified evaluation, cost logging, and open infrastructure for diagram-to-code agents.
    New benchmark artifact; independent evidence is the public HF release and verification scripts, not a physical entity.
  • Haiku→Sonnet→Opus cascade routing strategy for quantum circuit vision
    purpose: Reduce cost while recovering accuracy by escalating only failures.
    Instantiation of FrugalGPT-style cascading on this task; numbers are from this experiment’s failure sets, not a new theoretical object.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Circuit Vision: Cost-Aware Evaluation of Visual AI Agents for Quantum Code Generation." pith.science (2026). https://pith.science/paper/AZW23J5Z

@misc{pith2026260710057,
  author       = {Pith},
  title        = {Pith review of: Quantum Circuit Vision: Cost-Aware Evaluation of Visual AI Agents for Quantum Code Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AZW23J5Z}},
  note         = {Machine review of arXiv:2607.10057}
}
abstract

Can AI agents visually comprehend quantum circuit diagrams and generate verified executable code--and at what cost? We present Quantum Circuit Vision, a cost-aware evaluation framework for multimodal AI agents on quantum circuit visual understanding. We construct a 132-circuit benchmark spanning 13 categories ($1$--$10$ qubits) with executable Amazon Braket code and unitary-fidelity verification. Evaluating three frontier Claude-family models at different capability-cost tiers with $n=5$ repeated trials, we find that the mid-tier model (Sonnet 4.6, $1.30\times$ credits) offers the most favorable balance on the cost-accuracy frontier: 91% pass rate on the core subset at 18% of the per-call cost of the strongest model (Opus 4.6), whose accuracy advantage is not statistically significant (paired $t$: $p=0.083$). Logistic regression confirms that circuit depth--not qubit count--is the primary predictor of failure ($p<0.001$). Chain-of-thought prompting shows no statistically significant effect (all $p>0.18$, $n=5$), suggesting that visual pattern recognition outweighs explicit reasoning strategy for structurally coupled diagrams. We propose a cascade routing strategy (cheap $\rightarrow$ expensive models) that achieves 84% accuracy at 38% of single-model cost, demonstrating that model routing dominates prompt engineering as a cost lever. We release QCV-Dataset (132 circuits, 5 modalities, 1,931 files) on Hugging Face Hub as an open evaluation infrastructure with structured metadata for discoverability, interoperability, and responsible AI documentation, and all evaluation code, cost logs, and verification scripts on GitHub for full reproducibility.

Figures

Figures reproduced from arXiv: 2607.10057 by the authors.

Figure 1
Figure 1. QCV system overview. Top: QCV-Dataset provides 132 quantum circuits with five core data modalities—circuit images, [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Representative circuits. Left column (a–c): increasing complexity from trivial to challenging. Right (d): the largest [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Haiku on the Toffoli circuit: BV uses a nonexistent [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Eleven blockchain-relevant quantum circuits (2–8 qubits) spanning consensus, cryptography, threat modeling, key [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Proposed Hybrid Agent architecture (not yet imple [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 3 canonical work pages

  1. [1]

    Acharya et al

    R. Acharya et al. 2023. Suppressing quantum errors by scaling a surface code logical qubit.Nature614 (2023), 676–680

    2023

  2. [2]

    Mubashara Akhtar, Omar Benjelloun, Costanza Conforti, Luca Foschini, Pieter Gijsbers, Joan Giner-Miguelez, Sujata Goswami, Nitisha Jain, Michalis Karamou- sadakis, Satyapriya Krishna, Michael Kuchnik, Sylvain Lesage, Quentin Lhoest, Pierre Marcenac, Manil Maskey, Peter Mattson, Luis Oala, Hamidah Oderinwale, Pierre Ruyssen, Tim Santos, Rajat Shinde, Elena...

    doi:10.52202/079017-2610 2024

  3. [3]

    Farag, Taylor L

    Yuri Alexeev, Marwa H. Farag, Taylor L. Patti, Mark E. Wolf, Natalia Ares, Alán Aspuru-Guzik, Simon C. Benjamin, Zhenyu Cai, Shuxiang Cao, Christopher Chamberland, Zohim Chandani, Federico Fedele, Ikko Hamamura, Nicholas Harrigan, Jin-Sung Kim, Elica Kyoseva, Justin G. Lietz, Tom Lubowe, Alexander McCaskey, Roger G. Melko, Kouhei Nakaji, Alberto Peruzzo, ...

    doi:10.1038/s41467-025-65836-3 2025

  4. [4]

    Amazon Web Services. 2024. Amazon Braket Developer Guide. https://docs. aws.amazon.com/braket/

    2024

  5. [5]

    Arute et al

    F. Arute et al. 2019. Quantum supremacy using a programmable superconducting processor.Nature574 (2019), 505–510

    2019

  6. [6]

    Ville Bergholm et al . 2018. PennyLane: Automatic Differentiation of Hybrid Quantum-Classical Computations.arXiv preprint arXiv:1811.04968(2018). https: //arxiv.org/abs/1811.04968

    arXiv 2018

  7. [7]

    Blocklove, S

    J. Blocklove, S. Garg, R. Karri, and H. Pearce. 2023. Chip-Chat: Challenges and Opportunities in Conversational Hardware Design. InProc. ACM/IEEE MLCAD. 1–6

    2023

  8. [8]

    Lukas Burgholzer and Robert Wille. 2022. Handling Non-Unitaries in Quantum Circuit Equivalence Checking. InProc. ACM/IEEE DAC. 529–534. doi:10.1145/ 3489517.3530482

    arXiv 2022

Show all 50 references
  1. [9]

    Beysen- gulov, Daniel C

    Shuxiang Cao, Zijian Zhang, Abhishek Agarwal, Grace Bratrud, Niyaz R. Beysen- gulov, Daniel C. Cole, Alejandro Gómez Frieiro, Elena O. Glen, Hao Hsu, Gang Huang, Raymond Jow, Greshma Shaji, Tom Lubowe, Ligeng Zhu, Luis Mantilla Calderón, Nicola Pancotti, Joel Pendleton, Brando...

    2026 arXiv

  2. [10]

    Stephanie Russo Carroll, Edit Herczog, Maui Hudson, Keith Russell, and Shelley Stall. 2021. Operationalizing the CARE and FAIR Principles for Indigenous Data Futures.Scientific Data8 (2021), 108. doi:10.1038/s41597-021-00892-0

    2021 doi

  3. [11]

    Cerezo et al

    M. Cerezo et al. 2021. Variational quantum algorithms.Nature Reviews Physics3 (2021), 625–644

    2021

  4. [12]

    Ruiqi Chang, Daeun Kwon, Jihoon Lee, and Naveen Verma. 2026. CascadeDe- bate: Multi-Agent Deliberation for Cost-Aware LLM Cascades.arXiv preprint arXiv:2604.12262(2026). https://arxiv.org/abs/2604.12262

    2026 arXiv

  5. [13]

    Mohammad Chehimi and Walid Saad. 2022. Quantum Federated Learning with Quantum Data. InProc. IEEE ICASSP. 8617–8621. doi:10.1109/ICASSP43922.2022. 9746622

    2022 doi

  6. [14]

    Lingjiao Chen, Matei Zaharia, and James Zou. 2023. FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance.arXiv preprint arXiv:2305.05176(2023). https://arxiv.org/abs/2305.05176

    2023 arXiv

  7. [15]

    A. K. Fedorov, E. O. Kiktenko, and A. I. Lvovsky. 2018. Quantum Computers Put Blockchain Security at Risk.Nature563 (2018), 465–467

    2018

  8. [16]

    Fu et al

    Y. Fu et al. 2023. GPT4AIGChip: Towards Next-Generation AI Accelerator Design Automation via Large Language Models. InProc. ICCAD

    2023

  9. [17]

    Timnit Gebru, Jamie Morgenstern, Briana Vecchione, Jennifer Wortman Vaughan, Hanna Wallach, Hal Daumé III, and Kate Crawford. 2021. Datasheets for Datasets. Commun. ACM64, 12 (2021), 86–92

    2021

  10. [18]

    Albert Gu, Karan Goel, and Christopher Re. 2022. Efficiently Modeling Long Sequences with Structured State Spaces. InProc. ICLR

    2022

  11. [19]

    Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long Short-Term Memory. Neural Computation9, 8 (1997), 1735–1780

    1997

  12. [20]

    Jumper et al

    J. Jumper et al. 2021. Highly accurate protein structure prediction with AlphaFold. Nature596 (2021), 583–589

    2021

  13. [21]

    Kim et al

    Y. Kim et al. 2023. Evidence for the utility of quantum computing before fault tolerance.Nature618 (2023), 500–505

    2023

  14. [22]

    Kaixin Li, Yuchen Tian, Qisheng Hu, Ziyang Luo, and Jing Ma. 2024. MMCode: Benchmarking Multimodal Large Language Models for Code Generation with Visually Rich Programming Problems. InProc. EMNLP (Findings)

    2024

  15. [23]

    D. Liu, A. Zhang, and L. Zhang. 2026. Quantum-Safe, Efficient, and AI-Enhanced Blockchains for the Web. InCompanion Proc. ACM Web Conference (WWW’26)

    2026

  16. [24]

    H. Liu, C. Li, Q. Wu, and Y. J. Lee. 2023. Visual Instruction Tuning. InProc. NeurIPS

    2023

  17. [25]

    Liu et al

    M. Liu et al. 2023. ChipNemo: Domain-Adapted LLMs for Chip Design.arXiv preprint arXiv:2311.00176(2023)

    2023 arXiv

  18. [26]

    Shu Liu, Xin Yuan, Tian Chen, Zhiwei Zhan, Zhu Han, Da Zheng, et al . 2026. CASTER: Breaking the Cost-Performance Barrier in Multi-Agent Orchestra- tion via Context-Aware Strategy for Task Efficient Routing.arXiv preprint arXiv:2601.19793(2026). https://arxiv.org/abs/2601.19793

    2026

  19. [27]

    Lu et al

    P. Lu et al. 2023. MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts.arXiv preprint arXiv:2310.02255(2023)

    2023 arXiv

  20. [28]

    McKinsey & Company. 2024. Quantum Technology Monitor: How Quantum Computing Could Change the World. https://www.mckinsey.com/featured- insights/themes/how-quantum-computing-could-change-the-world Estimates $450–850B economic value by 2040

    2024

  21. [29]

    Kelleher

    Yasmin Moslem and John D. Kelleher. 2026. Dynamic Model Routing and Cas- cading for Efficient LLM Inference: A Survey.arXiv preprint arXiv:2603.04445 (2026). https://arxiv.org/abs/2603.04445

    2026 arXiv

  22. [30]

    Akira Nakayama, Kosuke Mitarai, Leonardo Placidi, Takanori Sugimoto, and Keisuke Fujii. 2025. VQE-Generated Quantum Circuit Dataset for Machine Learn- ing.Physical Review Research7, 3 (2025), 033021. doi:10.1103/PhysRevResearch. 7.033021

    2025 doi

  23. [31]

    M. A. Nielsen and I. L. Chuang. 2010.Quantum Computation and Quantum Information(10th anniversary ed.). Cambridge University Press

    2010

  24. [32]

    2023.GPT-4 Technical Report

    OpenAI. 2023.GPT-4 Technical Report. Technical Report

    2023

  25. [33]

    Elijah Perrier, Akram Youssry, and Chris Ferrie. 2022. QDataSet: Quantum Datasets for Machine Learning.Scientific Data9, 1 (2022), 582. doi:10.1038/s41597- 022-01639-1

    2022 doi

  26. [34]

    Leonardo Placidi, Ryuichiro Hataya, Toshio Mori, Koki Aoyama, Hayata Morisaki, Kosuke Mitarai, and Keisuke Fujii. 2026. MNISQ: A Large-Scale Quantum Circuit Dataset for Machine Learning in the NISQ Era.Scientific Data13 (2026), 810. doi:10.1038/s41597-026-07493-9

    2026 doi

  27. [35]

    Preskill

    J. Preskill. 2018. Quantum Computing in the NISQ era and beyond.Quantum2 (2018), 79

    2018

  28. [36]

    Cheng Qian, Zhibin Liu, Srajan Kokane, Akshat Prabhakar, Jiaxin Qiu, et al. 2025. xRouter: Training Cost-Aware LLMs Orchestration System via Reinforcement Learning.arXiv preprint arXiv:2510.08439(2025). https://arxiv.org/abs/2510. 08439

    2025

  29. [37]

    Qiskit Contributors. 2023. Qiskit: An Open-Source Framework for Quantum Computing

    2023

  30. [38]

    Coles, and M

    Louis Schatzki, Andrew Arrasmith, Patrick J. Coles, and M. Cerezo. 2021. Entan- gled Datasets for Quantum Machine Learning.arXiv preprint arXiv:2109.03400 (2021). https://arxiv.org/abs/2109.03400

    2021 arXiv

  31. [39]

    Reza Shirkavand, Shuzhi Gao, Philip Yu, et al . 2025. Cost-Aware Contrastive Routing for LLMs. InProc. NeurIPS

    2025

  32. [40]

    Shailja Thakur, Baleegh Ahmad, Hammond Pearce, Benjamin Tan, Brendan Dolan-Gavitt, Ramesh Karri, and Siddharth Garg. 2024. VeriGen: A Large Lan- guage Model for Verilog Code Generation.ACM Trans. Des. Autom. Electron. Syst.29, 3, Article 46 (April 2024), 31 pages. doi:10.1145/3643681

    2024 doi

  33. [41]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention Is All You Need. InProc. NeurIPS, Vol. 30

    2017

  34. [42]

    Viamontes, Igor L

    George F. Viamontes, Igor L. Markov, and John P. Hayes. 2007. Checking Equivalence of Quantum Circuits and States. InProc. IEEE/ACM ICCAD. 69–

    2007

  35. [43]

    doi:10.1109/ICCAD.2007.4397246

    2007 doi

  36. [44]

    Haoyu Wang, Xianglin Zhou, Zihan Xu, Kun Cheng, Yixin Zuo, Kai Tian, et al

  37. [45]

    https: //arxiv.org/abs/2502.11829

    CODE-VISION: Evaluating Multimodal LLMs Logic Understanding and Code Generation Capabilities.arXiv preprint arXiv:2502.11829(2025). https: //arxiv.org/abs/2502.11829

    2025 arXiv

  38. [46]

    Wilkinson, Michel Dumontier, IJsbrand Jan Aalbersberg, Gabrielle Apple- ton, Myles Axton, Arie Baak, Niklas Blomberg, Jan-Willem Boiten, Luiz Bonino da Silva Santos, Philip E

    Mark D. Wilkinson, Michel Dumontier, IJsbrand Jan Aalbersberg, Gabrielle Apple- ton, Myles Axton, Arie Baak, Niklas Blomberg, Jan-Willem Boiten, Luiz Bonino da Silva Santos, Philip E. Bourne, Jildau Bouwman, Anthony J. Brookes, Tim Clark, Mercè Crosas, Ingrid Dillo, Olivier Du...

    2016 doi

  39. [47]

    Wong, G.-W

    S.-Z. Wong, G.-W. Wan, D. Liu, and X. Wang. 2024. VGV: Verilog Generation using Visual Capabilities of Multi-Modal Large Language Models. InProc. IEEE/ACM DAC

    2024

  40. [48]

    Chengyue Wu, Zhenyu Liang, Yuchen Ge, Qi Guo, Zhiqiang Lu, Jing Wang, et al. 2025. Plot2Code: A Comprehensive Benchmark for Evaluating Multi-Modal Large Language Models in Code Generation from Scientific Plots. InProc. NAACL (Findings)

    2025

  41. [49]

    Haoyuan Wu, Zhuolun He, Xinyun Zhang, Xufeng Yao, Su Zheng, Haisheng Zheng, and Bei Yu. 2024. ChatEDA: A Large Language Model Powered Au- tonomous Agent for EDA.Trans. Comp.-Aided Des. Integ. Cir. Sys.43, 10 (Oct. 2024), 3184–3197. doi:10.1109/TCAD.2024.3383347

    2024 doi

  42. [50]

    Design 5q QKD

    Xinyang Zhao, Xianglin Luo, Qipeng Shi, Cheng Chen, Shuo Wang, et al. 2025. ChartCoder: Advancing Multimodal Large Language Model for Chart-to-Code Generation. InProc. ACL. A Blockchain Circuit Diagrams Figure 4 shows the eleven blockchain-relevant quantum circuits evaluated i...

    2025

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.