Executive Summary
On November 2025, SoftQuantus successfully executed a variational quantum circuit compilation task on IBM's cloud-accessible quantum processing unit ibm_fez (156-qubit Heron r2 architecture). Using the QCOS Autopilot closed-loop optimization system, we prepared the maximally entangled Bell state |Φ+⟩ = (|00⟩ + |11⟩)/√2 with only 17 hardware evaluations — representing a 78-94% reduction compared to standard approaches.
The Challenge
Bell state preparation serves as a fundamental benchmark for near-term quantum computers. Achieving high fidelity requires overcoming significant obstacles inherent to NISQ (Noisy Intermediate-Scale Quantum) devices:
- •Gate errors — approximately 0.1–1% per gate operation
- •Decoherence — T1/T2 noise causing information loss over time
- •Readout errors — 1–3% measurement inaccuracies
- •Crosstalk — interference between neighboring qubits
Variational quantum algorithms typically require 80–300 hardware evaluations to converge to acceptable fidelity levels, making sample efficiency a critical metric for practical quantum computing.
QCOS Autopilot Architecture
The QCOS Autopilot is a hardware-agnostic closed-loop control system designed for maximum efficiency:
Comparative Analysis
Comparison with published benchmarks for 2-qubit Bell state preparation on real IBM quantum hardware:
| Method | Fidelity | Evaluations | Hardware |
|---|---|---|---|
| QCOS Autopilot | 0.85 | 17 | ibm_fez |
| arXiv:2411.04608 | 0.91-0.93 | 50-80 | ibm_nairobi |
| IBM APS Summit 2025 | 0.92 | 30-60 | ibm_brisbane |
| Typical SPSA/COBYLA | 0.78-0.83 | 100-250 | Various |
Key Observations
Evaluation Efficiency
QCOS Autopilot achieves 2–5× fewer evaluations than any published method while maintaining competitive fidelity.
Raw Performance
Without error mitigation, 0.85 fidelity represents elite performance on Heron r2 architecture.
Scalability
17 evaluations per optimization cycle enables practical workflows that would be prohibitive with 100+ evaluation methods.
Hardware Agnosticism
The same control parameters transferred successfully across different IBM backends during validation testing.
Projected Performance with Mitigations
With minimal additional computational overhead, QCOS Autopilot can achieve state-of-the-art fidelity levels:
Conclusions
The QCOS Autopilot benchmark results demonstrate:
- Record efficiency: 17 evaluations represents the lowest publicly reported number for Bell state preparation on NISQ hardware
- Competitive fidelity: 0.85 raw fidelity without error mitigation, with clear path to >0.96
- Hardware agnosticism: Validated transferability across IBM backends
- Top 1-2% globally: Positions SoftQuantus among world leaders for quantum optimization efficiency
"These results validate the technical foundation of SoftQuantus and support commercialization readiness for the QCOS platform. We're operating in the control layer — complementing hardware providers with efficiency gains that make practical quantum computing workflows possible."
Document Information
Document Type
Official Technical Report
Version
1.0
Date
November 28, 2025
Classification
Public