Track
Compute
Maximize useful inference work per watt of electrical power.
Human baseline
~20 W brain-equivalent inference
Competition goal
Maximum useful work per watt (tokens/joule, FLOPS/watt)
Competition Announced
Submissions are not open yet. Join our Discord community for updates on tracks, timelines, and when entries open.
Human Mechanical Efficiency Competition
Design machines as efficient as humans.
A competition to design and build systems whose mechanical energy efficiency rivals the human body — across compute, mobility, and any domain that can be rigorously benchmarked.
The baseline
The human baseline
Every track in HMEComp measures your system against documented human energy efficiency. The question is simple: can your machine do the work using as little power as a person?
Biology solved the efficiency problem long before silicon or steel. HMEComp exists to push builders toward honest measurement — joules per unit of work, compared against a published human baseline.
“
Humans are the most energy-efficient general-purpose machines we know. Every watt counts — and comparing against the human baseline forces honest engineering.
— The north star metric
01
What
HMEComp is an open competition challenging engineers, researchers, and builders to achieve human-level mechanical energy efficiency in real systems.
02
Why
Humans are the most energy-efficient general-purpose machines we know. Every watt counts — and comparing against the human baseline forces honest engineering.
03
How
Pick a track, document your methodology, measure against a published human baseline, and submit reproducible results. Score = efficiency ratio × 100.
Competition tracks
Three paths to human-level efficiency
Compute, mobility, or bring your own benchmark in the open track. Each measures against a published human baseline.
Track
Mobility
Match or exceed human fuel-to-motion efficiency.
Human baseline
~25% walking efficiency, ~100 W sustained
Competition goal
Fuel or electricity-to-distance ratio vs. human locomotion
Track
Open
Any system where efficiency can be compared to humans.
Human baseline
Domain-specific, peer-documented human baseline
Competition goal
Demonstrate human-competitive efficiency in a novel domain
Reference data
What does human-level mean?
These reference benchmarks anchor every submission. Your efficiency score measures how close your system comes to matching them.
~100 W
Sustained metabolic power
Average power a human can sustain during moderate activity
~25%
Locomotion efficiency
Mechanical efficiency of human walking on level ground
~20 W
Brain compute power
Approximate power consumption of the human brain
~80 W
Basal metabolic rate
Resting energy expenditure for an average adult
Schedule
Timeline
The competition is announced. Submission dates will be confirmed on Discord as logistics finalize.
- 01Current
Announced
July 2026
Competition structure, tracks, and judging criteria published
- 02Upcoming
Submissions Open
Q3 2026
Entry portal opens; teams may register and submit benchmarks
- 03Upcoming
Judging
Q4 2026
Independent review of submitted methodologies and results
- 04Upcoming
Awards
Q4 2026
Winners announced per track and overall efficiency champion
Community
Join the community
Get track updates, discuss methodology, coordinate teams, and be first to know when submissions open.
discord.hmecomp.org