Built for purpose Not for scale
General-purpose AI has a hidden cost
The largest AI models contain up to 1.8 trillion parameters. Routing
every student submission through these systems carries a significant
environmental footprint.
Research shows the most energy-intensive models consume 33+ Wh
per long prompt over 70× more than smaller, purpose-built
alternatives.
Frontier models activate 37B+ parameters per query
Academic grading needs a fraction of that capacity
Over-engineering creates unnecessary emissions
GPT-4o estimated annual energy
390,000–460,000
Enough to power ~35,000 US homes
Energy per long prompt — frontier model
33+
Wh
70× higher than purpose-built alternatives
Model Architecture
A model built for one thing: assessment
Eduface Model
120B
total parameters
5%
Active at inference
6B
active per query
~5%
activation ratio
Sparse MoE architecture routes each task to only the relevant parameters, no wasted compute.
Frontier Models (est.)
~1.8T
total parameters (GPT-4, reported)
37%
Active at inference
37B+
active per query
15–20×
more compute
Designed for breadth across all knowledge domains — most of which are
irrelevant to academic assessment.
Mixtral 8×7B activates
13B / 47B
28% 3–5× more than Eduface
DeepSeek-V3 activates
37B / 671B
5.5% 6× more than Eduface
Eduface activates
6B / 120B
5% domain-optimised minimum
Sustainability data on request
We can provide energy consumption estimates per deployment, per course,
and per institution. Use our figures to support your institution's
sustainability reporting requirements.
Energy use per 1,000 student submissions
Estimated CO₂ equivalent vs. frontier models
ISO 14001-aligned environmental reporting
Annual estimate 10,000 student institution
Submissions processed
18 per student avg.
180,000
Energy use (Eduface)
3 Wh per submission
~540 kWh
Energy (frontier equivalent)
11× higher
~5,940 kWh
CO₂ saved vs. frontier
NL grid intensity 2025
~2.4 tonnes
Estimates based on published energy benchmarks for frontier models (Luccioni et al., 2023; Hugging Face, 2024) and internal Eduface inference measurements.
Sparse activation only what is needed
Step 01
Submission arrives
Step 02
Router activates experts
Our MoE architecture routes the task to only the 6B relevant parameters, skipping the other 114B entirely. Domain-specific experts handle assessment.
Step 03
Minimal compute, precise output
Feedback is generated using a fraction of the energy of a general-purpose model without compromising on quality or alignment to learning objectives.