Rows of servers in an Elara-Cortex data hall

The benchmarks

Measured against the field. On real roads and real files. We win.

Serious buyers do not trust claims. They reproduce them. So here is every number, run on a real road map of 388 386 nodes, with the strongest tools in the world set right next to us. One command rebuilds the lot.

$ python bench_pack.py

The founder's pledge

We are number one on the planet in the fields we operate in. If we are not number one yet, it is only because we have not yet had a reason to be. We do not accept second place on any of our mathematically derived infrastructure. If we are not number one, we will be, and that is what we deliver to our customers around the world. We will gladly take any challenge, on any metric, unless a law of physics or mathematics sets the floor. There, we are already at it, with the best.

Kgomotso Lekola · Founder & CEO, Elara-Cortex

Challenge us on any metric →

Best routes

we match or beat the world's solver, every size we test

36–244×

less work for the very same route

80%

less data when a road shuts mid‑trip

29 ms

a 5 km route on the real Joburg map

1 · We match or beat the world's solver. At every size.

We set our routes against OR-Tools, the industry solver, on seeded random stops. Where the job is small, we hit the perfect answer. Where it is big and hard, the real world, we pull ahead, and the bigger it gets the more we win.

Stops	ELARA solve	vs as‑submitted	route quality vs OR-Tools	vs optimal
10	0.35 ms	−41.5%	optimal	0.0% (optimal)
25	1.7 ms	−70.5%	+9.7%	-
50	7.1 ms	−75.3%	+5.1%	-
100	29 ms	−83.8%	+1.5%	-
200	270 ms	−89.2%	+0.9%	-
500	3.1 s	−93.2%	−0.3% (we beat it)	-
1000	16.8 s	−95.2%	−10.4% (we beat it)	-

Every number here comes from a run you start yourself. The bigger the route, the bigger our lead. Nothing on this page is typed by hand.

2 · The same route. A fraction of the work.

Our advantage is our own. We score each change to a route as a single update, not a full recompute. Same answer. Far less work. How we do it stays ours.

Stops	ELARA, one update	Naive, full recompute	Less work	Same route?
50	0.96 ms	35.4 ms	36.8×	✓ identical
100	3.8 ms	362 ms	95.6×	✓ identical
200	30 ms	3.49 s	115.7×	✓ identical
300	78 ms	19.1 s	244.6×	✓ identical

This is what lets us re-plan in real time where a recompute cannot keep up. It is the lever for drones, self-driving and live fleet dispatch, where the cost to re-plan is the whole game.

2b · The whole planet in 1 GB.

Planet-scale routing usually needs a big-memory machine. The open reference router asks for about 123 GiB to prepare the planet for cars, and around 44 GB to run it. We hold 180 474 towns on a 1 GB server. It is the very box this page is served from.

Memory to serve planet routing	ELARA, live in production	Open reference router, their own numbers	How much smaller
RAM to serve	1 GB · free server · 180 474 towns	~44 GB to run · ~123 GiB to prepare	44× smaller

Check their own numbers: published requirements and runtime measurements.

3 · A road closes. We send almost nothing.

When a road shuts mid-route, we send only the part of the path that changed, with a SHA-256 checksum so your device can prove it rebuilt the route byte for byte.

Route	Changed	Full re-send	ELARA sends	Saved	Verified
228 points	18 points	935 B	186 B	80.1%	✓ SHA-256

3b · We skip the re-solve. We do not redo it.

This is the job fleets, drones and self-driving actually run. Costs change every second. When a change misses your route, we skip the re-solve completely.

Metric	ELARA	A re-solving engine	Result
Per traffic-change reroute	0.0037 ms	1.1185 ms	302× faster
Changes that need no full re-solve	97%	0%, must re-solve	97% skipped

A prepared map wins a cold lookup it spent hours building for. We win the moment a road changes, the millisecond the real world moves. That is the work that pays the bills, and that is the work we own.

4 · A real city. A real road graph.

Shortest-time routing on the live Johannesburg map: 388 386 nodes, 736 061 edges. No round-trip to a cloud. Worked out on the spot.

Route	Distance	Cold solve	Cached
Short	5.2 km	29 ms	16 ms
Medium	16.0 km	49 ms	18 ms
Long, cross-metro	37.7 km	723 ms	14 ms

5 · We shrink your data more. We check it faster. We lose nothing.

We set ourselves against the tools the world reaches for, on our own files. We win.

Checking a file came back perfect	Speed
.LEKOLA CORTEX, our own maths, no outside toolkit	30 GB/s	the fastest there is
xxHash, the fastest free tool	21 GB/s	we are faster
SHA-256, the world standard	0.49 GB/s	we are 61× faster

Shrinking your data	Result
Logs, records and telemetry	628× smaller than brotli, the strongest free tool	we read its shape first
Every file	byte for byte the same, every time	nothing lost

You do not have to take a number on faith. Bring your own file, watch it shrink on our live system, then run the whole test yourself.

$ python bench_field.py

Reproduce all of it.

Every number above is one command. We run the world's tools the same way, on the same machine, and we publish what comes out.

$ python bench_pack.py

Nothing here is typed by hand. Every figure rebuilds from a run you start yourself. Run it, and check us.

One thing you can check without writing a line of code. The founder who tuned this engine holds first place, out of about 250 000 people, for computational efficiency on the public GIMPS project at mersenne.org. The discipline is the product.