Epoch Learn is the Free Compute Program for young innovators: free, energy-aware AI compute for students, classrooms, clubs, and young builders, admitted in cohorts and powered by the Eco Foundry engine. Every model runs wrapped with memory, curriculum context, policy, routing, and evaluation, so the smallest capable model can run safely closer to the learner.

Can I try Epoch Learn without an account?

Yes. The Open Model Playground is public and requires no sign-up. If your browser supports WebGPU, it runs a real small model (such as Qwen2.5 0.5B or Llama 3.2 1B) entirely in your browser via @mlc-ai/web-llm, wrapped by the Advanced Epoch Wrapper; otherwise it falls back to a clearly labelled deterministic preview. You can also explore the Laplace Learning Network curriculum graph.

Eco Foundry is the compute engine behind Epoch Learn, not a separate product. It power-caps hardware, schedules around clean energy, and routes every request to the lowest-power option that meets latency — a hybrid of on-device models and energy-efficient eco servers.

Is the compute really free?

Yes, for accepted cohorts. We admit classrooms, clubs, and student groups in waves so capacity is real, and the in-browser playground is free for everyone with no account. We keep costs near zero by routing to the smallest capable model, power-capping GPUs, and scheduling around clean energy.

Free, energy-aware AI compute — powered by Eco Foundry

Free compute foryoung innovators who build

Get free compute Open the playground

$0what an accepted cohort pays for compute

~5%of AI talent has the compute it needs (UNDP, 2024)

−13.7%GPU energy via power caps, measured (MIT Lincoln Lab)

We thank our supporters

What you get

Watts,
not invoices.

Four commitments, each with a number attached. The engine page cites every figure.

01

Free cohort compute

Accepted classrooms, clubs, and student cohorts get compute on the Eco Foundry engine at no cost. The public playground needs no account at all.

$0per cohort, per term

02

Energy-aware scheduling

Jobs are power-capped, shifted toward clean-grid windows, and placed on the lowest-carbon node that meets latency (Wiesner et al., 2022).

5–20%emissions cut

03

Structured tutoring

Every model runs inside the Epoch Wrapper: curriculum context, guardrails, evaluation, human review. Structure is why the same model helps instead of harms.

≈2×learning gains

04

Runs down to your browser

WebGPU first, edge next, eco servers only when needed. The smallest capable model, as close to the learner as possible.

0accounts required

Process

Join.Route.Learn.

The compute gap

The map is
the mission.

Compute is the new literacy, and it is hoarded. We route donated, power-capped hardware to the students the market skips.

~5%

of AI talent has the compute it needs — 1% on-prem, 4% renting by the hour (UNDP, 2024). The other 95% is who this program is for.

<2%of world datacenter capacity is in Africa (UNDP, 2025)

10–30×costlier GPUs relative to GDP per capita (UNDP, 2025)

online

FeynmanDGX Spark · GB10

online

LaplaceRTX 3080 Ti

edge

TuringM4 Max · edge

queued

Noether2× TITAN Xp

Fleet roster shown as configured — live status is simulated until the public dashboard ships.

SIMULATED

Energy is a
first-class budget.

Energy figures are labelled simulations of the engine's accounting — never billed, never canned as live. The clock is real.

Y = (ΔK + λ·R + μ·D) / (kWh + H)

Learning per watt

the objective Eco Foundry optimizes — knowledge gained per kWh and per hour of human review, not GPU utilization

at the active node

Grid carbon intensity

0 gCO₂/kWh

vs uncapped GPUs (MIT Lincoln Lab)

Energy via power caps

−0.0%

SmolLM2Qwen2.5Llama 3.2Gemini 2.5 Flashsmallest capable model, always

The engine

Powered by
Eco Foundry.

Six systems, one objective: the most learning per watt. Eco Foundry is the engine — never the headline.

Tutor

Edge Tutors

Small tutoring agents distilled to one subject, deployed to the nearest capable runtime.

Model

Curriculum Distillation

Compress a large general model into a curriculum-specific small model that keeps the pedagogy.

Router

Energy-Aware Inference

Route each request by latency, energy budget, carbon intensity, and pedagogical fit.

Infra

Cohort Sandboxes

Isolated model environments per class or study, observable without cross-contamination.

Privacy

Student Memory Vaults

Learning memory the student owns: off, session-only, or persistent under explicit permission.

Runtime

Browser / Edge Runtime

One ModelRuntime abstraction over WebGPU, WASM, edge workers, and the eco-server fallback.

WebGPUfirst

WASMfallback

≤300Wfleet power cap

Explore the engine→

Trust

Free,
not careless.

Young learners are the users. The constraints are stricter, not looser, because the compute is free.

Standing commitment

$0billed to students — simulated billing, never charged

No adsNo data resaleSim-labelled demosOpen methods

Privacy-first inference

Browser and edge runtimes keep learner data on the device wherever the task allows.

Transparent traces

Every turn exposes its routing, memory state, and evaluation — nothing hides behind a reply.

Student data boundaries

Memory is off, session-only, or a student-owned vault. Never silently retained.

Honest demos

Public demos are labelled simulated or experimental. Canned output is never presented as live.

Playground

Touch it first.
Trust it after.

The public playground runs real small models on your own hardware, before you make any account at all.

WebGPU in your tab

Small open models run entirely in the browser. Nothing leaves your device.

Wrapper trace, visible

Watch routing, context, and evaluation happen — the pedagogy is inspectable.

No account, no billing

The playground is public. Cohort compute is the only thing behind a waitlist.

Honest labelling

Everything simulated says so. Live capability checks read your real browser.

Open the playground Engine console

" " " " " " " "" "" """ " " " " " """" " """ " " " "" " " " "" " " " "" " " " " " " " " " " "" " "" " " "" """ """" "" " "" " " " " " " " "" " " "" " " "" " "" " "" " """ " "" " " " "" "" " " " " " " " " " """ """ " "" " " " " " " " " """ " " " " " "" " " " " " " "" " " " "" " "" " " " "" "" " " " " " " " "" "" " " " " " "" "" " " "" " " " " " "" " " """" """"" " " " " " " """ "" """ " " "" " " " " "" " " " " " " " " " " "" " " " "" " " " " " "" " " "" " "" " " " "" " " " " " " " " " " "" " " " " " " " "" " " " """ "" " " """ " " "" " " "" "" """ " "" "" " " "" " " "" " "" " " " " "" "" " "" " " " " " " " "" " "" " "" " """ " " "" " " " " " "" " " "" "" """ """ " "" " " " " " " """ " "" " """ " " " "" " " " " " " " " " " "" " " " " " " " " " " " " " " "" " "" "" "" " "" """ " "" "" """ " " " " " " " " "" " "" " " " " " " " " " "" "" "" " "" " " "" " """ "" " "" "" "" " " " " "" """" """ "" "" " " " "" " "" """ " """ "" " " " "" "" " " "" " "" " " """" " " " """" " " "" " " "" " "" " " " " " "" " " "" " """ " " " " " " " " "" "" " " " "" """ " " " "" " "" " " " "" " " " "" " " " " """ " " " "" " " " " " " " " " " "" "" " " " " "" """ "" "" "" " """ " " "" "" """" " " """ "" """ " " " " " " " "" """"" " " " " " " " " " " " " """ "" "" " " " " " " " "" " " " " """ " " "" " "" " " " " " "" " " " " "" "" " " """ " " """" " "" "" " "" " " "" """ " " " " " " "" " " " " " "" """ " " "" " """ """" " " """ " " " "" " """ " " " "" """ " " " " """"" " " "" " " " "" " "" """ " "" " " " " " " " " """ "" "" "" " "" " "" """ " " " "" " " "" "" "" " " " "" " "" " "" """ " " " " " " " " "" " "" " " " " " """" " " " "" " "" " " " " " " " " " " " " " " " " "" "" " " """ "" """ " " "" " """ " " "" " """" " "" " " "" "" " " " " "" " "" " " " " " " " " """ " " "" " " " " " " "" " " " " " " "" " " " " " " " " " " " " " """"" " "" " "" "" "" " " " " " " " " "" "" " " " " " " "" " " " """ " """ " " "" " " " " " """ " " " " " " " " " " " " "" " " " " " " " " """ " " " " " " " "" " " "" "" " " " " " " " """ " "" " " "" """ " """ """"" " "" " """ "" " " " " " " " """""" "" "" " " " "" " " " " " """ " " " " "" """ " "" """" " " " " " " " " """ " " "" "" "" "" " " " """ """" " """ " " "" """ " " " " " " " " " " """ " " " "" " " "" " " " " " " " " " "" " " " """ """ """ " """ " "" "" " " " " "" """ " " " " " " "" """" " " "" " " "" " "" " " " " "" " "" " " "" " " " " " " " "" " " "" " " " " " " " " " " " "" " " " " " " "" " " "" " " " " """ " " " " " " " " "" "" " " " " " " " " " """ " " """"" " " """ " " " " " "" " " " """ " """ " "" " "" " " " """ " " "" " " " " " "" "" """" " """ " " "" "" " " "" " " " " " " " "" " """ " " " " " " " " " " " " " " " " " "" " "" " " "" " "" " " " "" "" " " " " " " " " " " "" " "" " "" " " " " " "" " """ " " """""""" " "" "" " " " """ """ " " " " " """" " " " " " " " " " " " " " "" " " " " " """ "" " "" " "" "" """"" " " " " " " "" " " " " " "" " " " " "" "" " " " " """ " "" " " " " " " """ " " " " "

Evidence

The same model helps or harms. The difference is structure.

“

A well-structured AI tutor produced about twice the learning gains of in-class active learning — and 83% rated its explanations at or above their instructor’s.

K

Kestin et al.

Harvard RCT, n=194, 2025

≈2×learning gains, structured tutoring

Sources

Access

Pay with
intent.

Everything is $0. We keep costs near zero by routing to the smallest capable model, power-capping GPUs, and scheduling around clean energy — then give the capacity away in cohorts.

01

Playground

Public. No account, no billing.

$0

forever

In-browser small models (WebGPU)
Wrapper trace, fully visible
Curriculum graph explorer
Honest sim/live labelling

Open now

The program

02

Cohort Compute

For classrooms, clubs, and student cohorts.

$0

waitlist · admitted in cohorts

Eco-server compute on the Eco Foundry engine
Memory vaults + cohort sandboxes
Educator dashboard and evaluation traces
Real capacity, not a throttled trial

Join the waitlist

03

Partners & Research

Pilots, studies, and donated hardware.

Let's talk

Co-run programs and pilots
Research telemetry exports
Protocol evaluation at cohort scale
Energy accounting you can audit

Write to us

Simulated billing — never chargedSmallest capable model, alwaysClean-energy scheduling

Read how it stays free

Stop waiting
for compute.

Join the Free Compute Program waitlist. We admit classrooms, clubs, and student cohorts each term — and the playground is open right now.

$0 for accepted cohorts — simulated billing, never charged

Free compute foryoung innovators who build

Watts,not invoices.

Free cohort compute

Energy-aware scheduling

Structured tutoring

Runs down to your browser

Join.Route.Learn.

Join

Route

Learn

The map isthe mission.

Energy is afirst-class budget.

Powered byEco Foundry.

Free,not careless.

Privacy-first inference

Transparent traces

Student data boundaries

Honest demos

Touch it first.Trust it after.

WebGPU in your tab

Wrapper trace, visible

No account, no billing

Honest labelling

The same model helps or harms. The difference is structure.

Pay withintent.

Playground

Cohort Compute

Partners & Research

Stop waitingfor compute.

Watts,
not invoices.

The map is
the mission.

Energy is a
first-class budget.

Powered by
Eco Foundry.

Free,
not careless.

Touch it first.
Trust it after.

Pay with
intent.

Stop waiting
for compute.