Among route optimization APIs for parcel delivery, Kardinal leads this sector ranking by a wide margin over Solvice. Functional coverage remains the decisive factor: a handful of advanced constraints, such as real-time re-optimization, clearly separate the most complete vendors from the rest of the market.
Parcel delivery is one of the most demanding use cases in route optimization. High stop density, strict time windows, daily disruptions, constant cost pressure: last-mile operators have no room for error. Picking the wrong API can cost hours of manual rework every day, wiping out the theoretical gains of optimization altogether.
This article compares eleven route optimization APIs (Kardinal, Solvice, LogisticsOS, HERE, NextBillion AI, Google, GraphHopper, Timefold, Verso, Routific, Mapbox) on their ability to handle the real constraints of parcel delivery.
Key takeaways
What makes parcel delivery genuinely complex
Stop density
An urban delivery route typically groups 80 to 150 stops per driver per day. At that density, micro-decisions about sequencing have a direct impact on productivity: stop order, building access, repeat visits. Poor sequencing on a route that dense can waste 30 to 45 minutes per driver per day.
Time windows
Delivery slots, often one hour and sometimes thirty minutes, have become the e-commerce norm. A parcel delivered outside its window causes immediate customer dissatisfaction and usually a second attempt to arrange, which degrades profitability and service quality at the same time.
Mid-day disruptions
Parcel delivery is fundamentally unpredictable: a refused parcel, an absent recipient, a full pickup point, unexpected congestion. Every disruption affects the rest of the route. Without real-time re-optimization, meaning the engine's ability to automatically recalculate the remaining route when an event occurs, the dispatcher has to step in manually, several times a day. Four solutions cover this capability in full: Kardinal, Timefold, NextBillion AI and Solvice.
Last-minute orders
In advanced e-commerce operations, orders keep arriving until mid-morning for same-day delivery. The ability to inject new stops into routes already under way, without rebuilding them from scratch, is essential here, and overlaps heavily with the same constraint family as real-time re-optimization.
Cost pressure
The last mile accounts for 40 to 50% of the total logistics cost of an order. With margins below 5% in most e-commerce segments, every unnecessary kilometer and every unoptimized working hour weighs directly on profitability.
The critical constraints for a parcel delivery API
Three levels of requirement shape the market: the classic base that everyone covers, the intermediate constraints that separate the good from the average, and the advanced constraints that genuinely make the difference in the field.
Distance and working time
The fundamental objectives of any optimization.
Covered by: All vendors except Mapbox.
Strict time windows
Handling of multiple or nested windows varies widely from one vendor to the next.
Covered by: All vendors.
Vehicle capacity
Weight, volume and custom limits per vehicle.
Covered by: All vendors.
Retry and premium parcels
Ranking the previous day's failed deliveries ahead of today's orders.
Covered by: All vendors except Mapbox.
Predictive traffic upstream
Expected traffic is factored in before optimization, not corrected afterwards.
Covered by: Kardinal, Solvice, GraphHopper, Google, HERE, Mapbox, NextBillion AI. Partial at LogisticsOS; absent at Routific, Timefold and Verso.
Mixed speed profiles
Car, bike and van within the same fleet.
Covered by: Kardinal, NextBillion AI, GraphHopper and HERE. Partial elsewhere.
Preferred zone assignment
Steering deliveries toward the courier who knows the area, without making it a hard rule.
Covered by: Kardinal, Timefold, Solvice, GraphHopper, HERE and LogisticsOS.
Workload balancing
Spreading working time fairly across couriers.
Covered by: Kardinal, Routific, NextBillion AI, Solvice, GraphHopper and LogisticsOS.
Real-time re-optimization
Automatic recalculation of the route during the day.
Covered by: Kardinal, Timefold, NextBillion AI and Solvice. Partial at Routific, Verso and LogisticsOS.
Lexicographic optimization
Ranking objectives instead of weighting them into a single formula.
Covered by: Kardinal and HERE only.
Route compactness
Keeping each route geographically tight and limiting overlap between couriers.
Covered by: Kardinal, Solvice and LogisticsOS only.
Zone-by-zone sequencing
Finishing one neighborhood before moving to the next.
Covered by: Kardinal, Solvice, GraphHopper and HERE.
Conditional return pickups
Only starting a return collection once the route is far enough along, without excessive rigidity.
Covered by: Kardinal and Solvice only.
Best-effort optional stops
Scheduling flexible-deadline parcels only when they fit without adding a vehicle.
Covered by: Kardinal, Timefold, NextBillion AI and Solvice.
Pickup-point anchoring
Forcing the service order around a pickup point (PUDO), a niche constraint.
Covered by: HERE only, ahead of Kardinal included.
Parcel delivery API comparison
Kardinal
95% coverage · 1st/11Kardinal remains the best-suited solution for parcel delivery, whatever the complexity segment. It is the only vendor to combine real-time re-optimization, route compactness, zone-by-zone sequencing and lexicographic optimization in a single offering.
Across all the constraints specific to this sector, Kardinal has only two gaps: hourly cost that varies with the time of day, a morning-peak surcharge for instance, which only Solvice models, and tying stops to a pickup point, which only HERE offers.
Solvice
79% coverage · 2nd/11Solvice takes second place and stands out with one advantage Kardinal does not cover: modeling hourly cost according to the time of day. It also joins Kardinal on conditional return pickups, a constraint these two solutions alone cover.
Solvice also fully covers real-time re-optimization, route compactness and zone-by-zone sequencing.
LogisticsOS
77% coverage · 3rd/11LogisticsOS lands straight in third place, carried by its specialty: cost modeling. It is the only solution, alongside Kardinal on a neighboring axis, to cover flat-rate and overtime thresholds as well as optimizing the combined bill across own fleet and subcontractors. It also covers route compactness.
Its limits: real-time re-optimization and zone assignment are only partially covered.
HERE
73% coverage · 4th/11HERE surprises with two exclusives. It is the only solution besides Kardinal to offer lexicographic optimization, and the only one in the panel, Kardinal included, to cover pickup-point anchoring. It also leverages its mapping base for solid predictive traffic and complete speed profiles.
Its main weakness: real-time re-optimization is not covered, which penalizes it on operations with a high volume of daily disruptions.
NextBillion AI
68% coverage · 5th/11NextBillion AI remains solid on the fundamentals: predictive traffic, speed profiles, real-time re-optimization and best-effort optional stops are all covered. What it lacks is the finest layer: no route compactness, no zone-by-zone sequencing, no lexicographic optimization.
Google does not fully cover real-time re-optimization within this constraint scope, despite the quality of its mapping infrastructure. Predictive traffic and minimizing delays against a promised ETA remain strengths, but specialized speed profiles (bike, motorcycle) are absent, and no zone assignment is offered.
GraphHopper
52% coverage · 7th/11GraphHopper holds its position thanks to its mixed speed profiles and zone-by-zone sequencing, two points where it joins Kardinal, Solvice and HERE. It covers neither real-time re-optimization, nor lexicographic optimization, nor optional stops.
Timefold
48% coverage · 8th/11Timefold surprises by fully covering real-time re-optimization and best-effort optional stops, two advanced constraints where it joins Kardinal. But it has no predictive traffic, and its coverage of the basic constraints lags behind the more generalist vendors.
Verso, Routific and Mapbox
44% · 38% · 26%These three solutions hit their limits as soon as a fleet exceeds about ten vehicles or operations require fine-grained disruption handling.
- Verso (44%) stays limited to basic operations, without complete real-time re-optimization.
- Routific (38%) covers the fundamentals but lacks most advanced capabilities.
- Mapbox (26%) benefits from good predictive traffic thanks to its mapping base, but remains above all a visualization tool rather than a genuine route optimizer.
Summary table
| Solution | Coverage | Real time | Predictive traffic | Speed profiles | Lexicographic |
|---|---|---|---|---|---|
| Kardinal | 95% | Full | Full | Full | Full |
| Solvice | 79% | Full | Full | Partial | No |
| LogisticsOS | 77% | Partial | Partial | Partial | No |
| HERE | 73% | No | Full | Full | Full |
| NextBillion AI | 68% | Full | Full | Full | No |
| 66% | No | Full | Partial | No | |
| GraphHopper | 52% | No | Full | Full | No |
| Timefold | 48% | Full | No | Partial | No |
| Verso | 44% | Partial | No | Partial | No |
| Routific | 38% | Partial | No | Partial | No |
| Mapbox | 26% | No | Full | Partial | No |
Our verdict by operation size
Create a free account and get your first optimized parcel routes in minutes, on your real data, with no commitment.
FAQ
Which route optimization API should you choose for last-mile delivery?
Kardinal is the most complete solution for last-mile delivery, with 95% coverage of the sector's constraints. Solvice follows in second place (79%), ahead of LogisticsOS (77%) and HERE (73%). Entry-level solutions such as Routific or Verso remain suitable for small fleets without advanced constraints.
How does a route optimization API handle undelivered parcels?
Handling undelivered parcels relies on real-time re-optimization: the ability to automatically recalculate the remaining route when a parcel is not delivered, either reinserting the stop at the end of the route or redirecting it to a pickup point. Kardinal, Timefold, NextBillion AI and Solvice cover this capability in full. Other solutions require the dispatcher to step in manually.
Does Google cover real-time re-optimization?
No, not in full. Within the constraint scope analyzed, Kardinal, Timefold, NextBillion AI and Solvice cover real-time re-optimization; Google covers it only partially.
What is lexicographic optimization and why does it matter for parcel delivery?
Lexicographic optimization means setting an absolute hierarchy between objectives instead of combining them into a weighted formula, for example: first minimize time-window lateness, then, at equal service quality, minimize total distance. The approach is more robust than classic weighting, which often produces side effects that are hard to anticipate at scale. Only Kardinal and HERE offer it.
Do route optimization APIs handle mixed fleets of cargo bikes and vans?
Multi-modal fleet handling varies by solution. Kardinal, NextBillion AI, GraphHopper and HERE fully cover mixed speed profiles within a single optimization. The other solutions handle them only partially.
How do you choose between a free and a paid route optimization API?
There is no truly free route optimization API for professional use. GraphHopper, open source by origin, offers a self-hosted version, as does the Timefold solver in its Community Edition, but deploying and maintaining them carries a real cost. To compare the pricing of the eleven solutions in this guide, see our 2026 pricing comparison. For professional parcel delivery, the cost of the API is generally far outweighed by the operational gains: a 10 to 15% reduction in kilometers driven across a 20-vehicle fleet is worth tens of thousands of euros a year.
Going further
Retail & distribution
Store replenishment, receiving windows and multi-depot rotations: the constraints that separate the APIs in distribution.
Read the full analysis →
Oversized & heavy freight
Vehicle dimensions, restricted access and multi-dimensional capacity: the most discriminating sector in the benchmark.
Read the full analysis →
Long-haul / FTL
Regulation, driving hours and rate modeling on full truckload.
Read the full analysis →