v0.9.3LatestApril 21, 2026
- NewQAOA solver now supports the Parity Twine Chain (PTC) ansatz — dense-QUBO two-qubit depth drops from O(n²) to O(n). Enable with ansatz="parity_twine" on QAOASolver. Read the release note: /blog/parity-twine.
- NewResource estimator API: packages.solvers.parity_twine exposes brick_wall_pairs, ptc_gate_list, and count_ptc_resources for circuit-depth planning without CUDA-Q installed.
- Newscripts/bench_parity_twine.py ships CX-count and two-qubit-depth benchmarks across n ∈ {4, 8, 16, 32, 64, 120} with schedule-coverage assertions.
- ImprovedQAOA solver now emits a warning when invoked on a dense QUBO (density > 0.5, n ≥ 10) with the linear ansatz, suggesting the PTC alternative.
- ImprovedQAOASolver result metadata now includes the ansatz field for experiment reproducibility.
v0.9.2PriorApril 14, 2026
- NewMulti-GPU Annealing now supports up to 8× A10G cards with automatic population-split distribution.
- NewWebSocket streaming now emits per-iteration energy trace for QAOA depth sweeps.
- ImprovedGPU Annealing cooling schedule is now adaptive per-chain — average solution quality improved 8% on TSPLIB benchmarks.
- ImprovedPython SDK async client now uses httpx instead of aiohttp — 30% faster job submission under concurrent load.
- FixFixed edge case where Hybrid Solver would hang on disconnected graph inputs.
- FixCorrected GPU-second billing for jobs interrupted by timeout — only elapsed seconds are billed.
v0.9.0MajorMarch 28, 2026
- NewQAOA solver now supports arbitrary problem Hamiltonians via custom operator definitions.
- NewSimulation Sandbox: drag-and-drop workflow builder for chaining multiple solvers on the same problem.
- NewDataset Hub launched with 120+ benchmark instances from TSPLIB, QPLIB, and OR-Library.
- NewExperiment tracker stores full run history with reproducible parameter snapshots.
- ImprovedVQE optimizer upgraded to ADAM with gradient clipping — 2× faster convergence on Ising Hamiltonians.
- ImprovedREST API latency reduced by ~40ms through connection pooling at the GPU dispatch layer.
- FixPortfolio QUBO mapper now correctly handles negative correlations in covariance matrices.
v0.8.5PatchMarch 10, 2026
- NewAdded Bin Packing QUBO mapper supporting 1D, 2D, and 3D container variants.
- NewJob history now exportable as JSON or CSV from the dashboard.
- ImprovedTabu Search tenure now auto-tuned based on problem size — no manual configuration needed for most cases.
- FixResolved rate limiter regression introduced in v0.8.4 that incorrectly throttled Pro-tier API calls.
- FixQUBO matrix validation now catches complex-valued entries with a clear error message.
v0.8.0MajorFebruary 20, 2026
- NewVQE solver shipped — variational quantum eigensolver for Ising Hamiltonian ground state problems.
- NewCommunity Detection QUBO mapper added (modularity maximization).
- NewPython SDK: streaming iterator now supports async for-loops.
- ImprovedGPU Annealing now runs 4,096 parallel chains (up from 2,048) on A10G hardware.
- ImprovedAPI authentication now supports scoped API keys per team member (Business plan).
v0.7.0LaunchJanuary 28, 2026
- NewNEROX public beta launched. GPU Parallel Annealing, Simulated Annealing, Tabu Search, Hybrid Solver, and QAOA available.
- NewREST API and Python SDK (v0.1.0) shipped.
- NewTSP, CVRP, Portfolio, Job Scheduling, and MaxCut QUBO mappers available out of the box.
- NewBenchmark Hub launched with initial leaderboard for TSP-500 and MaxCut-200 instances.
Nerox
GPU-native quantum-inspired optimization for combinatorial problems at any scale.
© 2026 NEROX | All rights reserved.