Scientific problems, solved with the conditions written down
QCOS runs quantum algorithms for chemistry, optimisation, dynamics and error correction. Every result on this page states what was measured, on which backend and how far it goes, and every runtime call leaves a sealed record you can check.
- Interfaces
- REST API, Python SDK, CLI and MCP
- Backends
- Four simulation methods
- Evidence
- Every runtime call sealed
Problems and capabilities
What QCOS solves today
All capabilities below are available through the QCOS simulation API. Results come from simulators.
Molecular ground-state energy
The variational quantum eigensolver (VQE) reaches chemical accuracy on two-qubit molecular models and improves on Hartree-Fock for H4.
Combinatorial optimisation
QAOA reproduces the 0.6924 approximation ratio for depth-1 MaxCut on 3-regular graphs, the reference value from the original QAOA analysis.
Portfolio selection
Portfolio choices written as a QUBO and solved with a tested quantum optimisation path.
Quantum dynamics
Trotterised time evolution of Ising models, for studying how quantum systems change over time.
Quantum error correction
Surface codes at distance 3 and 5, decoded with belief propagation (BP), ordered-statistics decoding (OSD) and BP+OSD.
Error-aware qubit routing
A median SWAP ratio of 0.971 against the SABRE baseline over 630 paired circuits on synthetic topologies. A modest gain, stated with its statistics on the results page.
Factoring demonstration
Shor's algorithm factors 15. A textbook demonstration of the method, not a threat to RSA keys.
Noise and hardware tooling
Zero-noise extrapolation on a synthetic noise model, calibration fits, pulse-level tools and resource estimation.
Algorithms in benchmarking
Grover search, the quantum Fourier transform, phase estimation and quantum machine learning are implemented. Their results will appear here when they pass preregistered tests.
Simulation backends
Four methods, each with a stated ceiling
A request beyond a ceiling is refused before memory is allocated. Results are never silently truncated.
| Method | Use it for | Ceiling |
|---|---|---|
| State vector | Exact amplitudes of small circuits | 24 qubits |
| Density matrix | Circuits with noise channels | 10 qubits |
| Clifford (stabiliser) | Large error-correction circuits | 24 qubits by default; Clifford gates only |
| Matrix product state | Weakly entangled circuits | Bond dimension 256 and a time limit; refused, not truncated |
Interfaces
One API, four ways in
QCOS is defined once in an OpenAPI contract and reached through the interface that fits your work, from a notebook to a CI pipeline to an AI agent.
REST API
293 documented paths.
Python SDK and CLI
The softqcos package on PyPI.
MCP server
softqcos-mcp gives AI agents 140 tools over the same contract.
Evidence
Every runtime call leaves a record sealed under a SHA-256 Merkle root.
Questions
What buyers ask about QCOS algorithms
Do these results come from quantum hardware?
No. They come from simulators, and the execution_mode label on every result says so. Access to quantum hardware through Softquantus Cloud is in development.
Do these results show an advantage over classical computers?
No such claim is made. The results show that QCOS reproduces known reference values under stated conditions.
Can I reproduce a result?
Yes. Every runtime call leaves a sealed record whose hashes you can recompute, and our benchmarks are preregistered and hash-pinned.
Can AI agents run experiments?
Yes. The softqcos-mcp server exposes 140 tools to MCP clients. See MCP integration.
Bring a problem from your lab
Tell us the molecule, the graph or the code distance. We will show what QCOS runs today and where its limits are.