---
title: Garmo Solvers — comparative evidence | SpinForge
url: https://spin-forge.com/results/garmo-solvers
description: A reported win, a loss, a correct-but-slower result and no recorded comparison are different outcomes. The records below retain those differences.
---

GARMO SOLVERS 1.0.2 / SOURCE-REPORTED SPINFORGE / A GARMO LABS COMPANY

# Comparisons you can
read precisely.

A reported win, a loss, a correct-but-slower result and no recorded comparison are different outcomes. The records below retain those differences.

[Explore the 21 modules](https://spin-forge.com/solvers) [Talk about your workload](https://spin-forge.com/contact)

RETAINED COMPARATIVE VERDICTS

## Useful evidence.
No universal ranking.

| Module | Standing | Reported comparison |

| MaxSAT | Reported win | Reported win against NuWLS-c-IBR, including a separate held-out result. Slash-separated study outcomes remain in the source; their unlabeled ordering is not reinterpreted. |

| CertCut | Reported win | 67 versus 49 proved instances against HiGHS over all 148 Biq Mac instances; zero mismatches in the reported comparison. |

| CP | Reported scheduling win | Reported scheduling win against CP-SAT 9.15. Broader XCSP coverage: 57/60 parsed and 19/60 decided. |

| Partition | Reported win | Reported win against mt-KaHyPar at a 60-second budget. Keep dataset and outcome details attached to the comparison. |

| VRP | Reported narrow win | Reported narrow win against PyVRP on the full Solomon set, not universal superiority on all routing workloads. |

| SMT | Reported scale win | 20/20 certified; 41.4 seconds versus 52.4 seconds against Z3 in the reported scale study. Z3 was faster on the sub-second cases. |

| Router | Reported portfolio win | Reported regret 0.046 versus 0.190 for the best single choice in the study. The source also reports a GPU/CPU portfolio comparison. |

| QBF | Reported loss | 15 versus 18 against DepQBF on 100 never-used inputs; the source notes missed true Hex games. |

| TSP | Reported loss | The source reports a held-out loss against LKH-3, with 2.0–3.8× gaps on the loss cases. No blanket advantage is claimed. |

| Presolve | Not promoted | Held-out benefit unconfirmed. The documented default keeps this path off. |

| Conic | No GPU gain | Clarabel solved the most in the reported comparison; the GPU path did not help. |

| CertOT | Correct; slower | Correct but slower than SciPy on all ten reported comparison cases. |

| MINLP | Completion evidence | 10/10 reported optimal results; the source makes no algorithmic advantage claim. |

| CertBQP | No comparison recorded | The supplied release summary records the capability but no head-to-head verdict. |

| certnum | No comparison recorded | The source identifies MDP error bounds, AIS partition-function bounds and model counting checked against OEIS; no head-to-head verdict is recorded. |

| MIP | No later verdict | An early audit was worse on 28/91 easy inputs; a time-limit defect was fixed. The supplied summary records no later comparison. |

| Krylov | Comparison incomplete | Compared with CuPy, Torch and the checksummed Forge15 old path; the summarized improvement pass was not started. |

| HOBO | No comparison recorded | Capability recorded; no head-to-head result in the supplied summary. |

| Toroidal | No comparison recorded | The source describes proving toroidal Gset instances, without a recorded external head-to-head verdict. |

| Batch | No comparison recorded | Capability recorded; no comparative result is supplied in this release account. |

| Unroute | No comparison recorded | Cut-certificate capability recorded; no external comparative verdict in the supplied summary. |

READ THE RESULT CORRECTLY

## One record.
One declared comparison.

The owner’s source defines standing as its latest measured result against an external baseline at equal time. Unlabeled slash-separated triples are not relabeled here. The study reports are not independently rerun for this publication. Full benchmark conditions belong in evaluation materials.

Garmo Solvers Router’s reported portfolio advantage does not overwrite UNISOLVE’s separate negative learned-routing findings. Garmo Solvers 1.0.2 also records MaxSAT changes and the TSP move rebuild; 1.0.1 turned QBF preprocessing off.