INTRODUCING ATHENA · MVRKO-SIM-1

A model built to predict what happens next.

Athena is a 4B Large Event Model that interprets sequences of real-world events, learns how behavior unfolds over time, and predicts the next likely event so products can make better decisions.

Markopolo AI automatically sends 1:1 hyper personalized emails to each shopper who leaves without buying.

Get early access

THE FIRST ATHENA RELEASE

We trained Athena on shopper journeys. But its
event-modeling approach extends far beyond eCommerce.

We trained Athena on shopper journeys. But its event-modeling approach extends far beyond eCommerce.

Athena’s first release was developed for eCommerce shopper behavior. Given the page a shopper sees, their interaction history, and their goal, it predicts the exact action they are most likely to take next.

A Real Shopping Journey

The model sees more than a final click.

It receives the current page state together with the journey that led there and the shopper’s intended outcome.

CURRENT PAGE

Running-shoe product detail page

HISTORY

Searched → compared → opened reviews

SHOPPER GOAL

Find a suitable pair under $120

From Browser to Traning Target

A specialist built around the structure of behavior.

01 · OBSERVE

Capture the rendered page, interaction history, current goal, and elements the shopper can act on.

02 · REDUCE

Remove styling and irrelevant markup while preserving actionable elements, hierarchy, function, and stable semantic IDs.

03 · SPECIALIZE

Fine-tune Qwen3-4B with LoRA over a 32K context window using the shopper’s next action as the target.

04 · EMIT

Return a deterministic structured action: action type, target, semantic ID, and input text when required.

Qwen3-4B

open base model

LoRA

specialist adaptation

32,768 tokens

long journey context

Structured output

exact next-action target

eCommerce is the first proving ground, not the limit.

The same event-modeling pattern can apply wherever behavior can be represented as an ordered sequence.

WHY ATHENA EXISTS

Language models understand words. Athena understands sequences.

Athena, unlike general language models, is designed to reason over what happened, in what order, and what is likely to happen next.

Observed events

What has happened

Athena

Interprets the sequence

Next-event prediction

What is likely next

Application decision

What the product should do

BENCHMARK PROOF

Specialized for next-action prediction

Athena ranked first in our full OPeRA next-action evaluation.

Strict exact match on OPeRA-test

Athena · mvrko-sim-1

24.50%

GPT-5.6

22.58%

GPT-4.1

21.50%

Claude Sonnet 5

18.35%

Evaluation setup

All internally tested models used the same 992 examples, parser, harness, and strict exact-match scorer.

Important caveat

Athena was specialist-fine-tuned, while frontier models were evaluated zero-shot. Its 1.92-point lead over GPT-5.6 is within sampling error.

HOW ATHENA WORKS

From event history to a structured next-event prediction

From event history to a structured
next-event prediction

From event history to a structured next-event prediction

Athena converts a sequence of observations into a prediction that applications can use directly.

CROSS-INDUSTRY APPLICATIONS

One event model pattern. Many possible systems.

Athena starts with next-action prediction, but the same event-sequence approach can support products wherever meaningful behavior unfolds over time.

Fraud and risk

Detect suspicious behavioral sequences and anticipate risky next actions.

Mobile products

Predict onboarding drop-off, return behavior, or likely user actions.

Education

Interpret learning sequences and predict where a student needs support.

Customer retention

Identify churn patterns and determine relevant interventions.

Conversational and voice agents

Use journey history to decide what an agent should say or do next.

Gaming

Model player progression, live-operations behavior, and in-game decisions.

B2B and sales

Understand account-event trails and likely buying progression.

Research and developer tooling

Create benchmarks, event converters, evaluation interfaces, and simulation systems.

Fraud and risk

Detect suspicious behavioral sequences and anticipate risky next actions.

Mobile products

Predict onboarding drop-off, return behavior, or likely user actions.

Education

Interpret learning sequences and predict where a student needs support.

Customer retention

Identify churn patterns and determine relevant interventions.

Conversational and voice agents

Use journey history to decide what an agent should say or do next.

Gaming

Model player progression, live-operations behavior, and in-game decisions.

B2B and sales

Understand account-event trails and likely buying progression.

Research and developer tooling

Create benchmarks, event converters, evaluation interfaces, and simulation systems.

Fraud and risk

Detect suspicious behavioral sequences and anticipate risky next actions.

Conversational and voice agents

Use journey history to decide what an agent should say or do next.

Mobile products

Predict onboarding drop-off, return behavior, or likely user actions.

Gaming

Model player progression, live-operations behavior, and in-game decisions.

Education

Interpret learning sequences and predict where a student needs support.

B2B and sales

Understand account-event trails and likely buying progression.

Customer retention

Identify churn patterns and determine relevant interventions.

Research and developer tooling

Create benchmarks, event converters, evaluation interfaces, and simulation systems.

These are starting points.

Athena can be tested anywhere meaningful behavior can be represented as a sequence of events. This is where hackathon-generated use cases eventually become valuable.

BUILDERS & UNIVERSITY TEAMS

Help discover what Athena can become

We are giving selected university teams early access to Athena to explore applications across industries.

Join the hackathon waitlist

ACCESS & TOOLING

Try and build with Athena

We trained Athena on shopper journeys. But its event-modeling approach extends far beyond eCommerce.

Choose the route that matches what you need to test: inspect the model, request a deployment package, or reproduce the evaluation setup.

Choose an access route

Use the public model page or request approved early access.

Represent your events

Convert meaningful behavior into a structured, ordered event sequence.

Evaluate predictions

Compare predicted next events with the exact actions in your test set.

ROUTES WE CAN PUBLISH

Start with the paths that are actually available.

Hugging Face model page

Inspect the model release and model-card details.

To confirm

Request early access

Apply for selected-team, design-partner, or managed-evaluation access.

Available

Self-hosted model package

Approved partners can evaluate deployment with data kept in their environment.

Early access

Benchmark resources

Inspect the OPeRA methodology, dataset, examples, and evaluation setup.

Public route

PUBLISH ONLY WHEN LIVE

Access routes still need launch confirmation.

Keep these in the wireframe as a readiness checklist, not as public buttons, until each destination exists.

Hosted playground

Unconfirmed

API access

Pilot route to confirm

Downloadable weights or adapter

Partner package

Starter notebook

Unconfirmed

Documentation

Access package to confirm

Example event schemas

Access package to confirm

Evaluation toolkit

Reproduction package to confirm

GitHub repository

Unconfirmed

ACCEEARLY ACCESSSS & TOOLING

Build the next decision system on a
model that understands events.

We trained Athena on shopper journeys. But its event-modeling approach extends far beyond eCommerce.

Join the Athena design-partner program to evaluate next-event

prediction on your own event streams and application workflows.

Join the Athena design-partner program to evaluate next-event

prediction on your own event streams and application workflows.

Get early access

No credit card required.

Join Our Community

Join Our Community

Join Our Community

Join Our Community

A model built to predict what happens next.

A model built to predict what happens next.

We trained Athena on shopper journeys. But itsevent-modeling approach extends far beyond eCommerce.

We trained Athena on shopper journeys. But its event-modeling approach extends far beyond eCommerce.

We trained Athena on shopper journeys. But its event-modeling approach extends far beyond eCommerce.

eCommerce is the first proving ground, not the limit.

eCommerce is the first proving ground, not the limit.

Language models understand words. Athena understands sequences.

Language models understand words. Athena understands sequences.

Specialized for next-action prediction

Specialized for next-action prediction

From event history to a structured next-event prediction

From event history to a structurednext-event prediction

From event history to a structured next-event prediction

One event model pattern. Many possible systems.

One event model pattern. Many possible systems.

These are starting points.

These are starting points.

Help discover what Athena can become

Help discover what Athena can become

Try and build with Athena

We trained Athena on shopper journeys. But its event-modeling approach extends far beyond eCommerce.

We trained Athena on shopper journeys. But its event-modeling approach extends far beyond eCommerce.

Build the next decision system on amodel that understands events.

We trained Athena on shopper journeys. But its event-modeling approach extends far beyond eCommerce.

We trained Athena on shopper journeys. But its event-modeling approach extends far beyond eCommerce.

We trained Athena on shopper journeys. But its
event-modeling approach extends far beyond eCommerce.

From event history to a structured
next-event prediction

Build the next decision system on a
model that understands events.