---
title: Mythos Benchmark — AI Model Complexity Tier for Frontier Research
description: Mythos and Capybara are Cromus's top-tier AI model complexity labels, used in Workflow Classification to identify tasks requiring maximum reasoning capability, specialized research, or custom fine-tuning beyond standard Frontier models.
canonical: https://cromus.ai/mythos-benchmark
source_html: https://cromus.ai/mythos-benchmark
---

# Mythos Benchmark

> The top-tier AI model complexity label in the Cromus Workflow Classification system.

In the Cromus 4-step measurement framework, **Workflow Classification** assigns every AI task to one of four complexity tiers. **Mythos** (and its paired label **Capybara**) is the highest tier — reserved for tasks that exceed the capability profile of standard Frontier models.

---

## The four tiers

| Tier | Complexity profile | Typical use |
|------|-------------------|-------------|
| **Lightweight** | Simple, structured tasks | Classification, extraction, routing |
| **Balanced** | Moderate reasoning | Drafting, summarization, multi-step |
| **Frontier** | Advanced reasoning | Code generation, complex planning |
| **Mythos / Capybara** | Maximum capability | Specialized research, scientific reasoning, custom fine-tunes |

---

## What Mythos / Capybara means in practice

Assigning a task to the Mythos/Capybara tier signals:

- The task requires top-tier reasoning that Frontier models handle inconsistently
- The output quality justifies the highest per-token cost in the registry
- Standard cost-quality modes (Eco, Cost, Balanced, Quality) are insufficient; only the top Quality mode or a custom fine-tune applies
- TCWO for this workflow is substantially higher than Frontier-tier equivalents

In the Cromus cost simulator, tasks classified as Mythos/Capybara are modeled against the highest-cost models in the registry and flagged explicitly in the Croms™ analysis — because the potential for optimization (e.g., decomposing the task into Frontier + Lightweight sub-tasks) is highest here.

---

## Croms™ and the Mythos tier

Tasks at the Mythos tier often contain the highest concentration of Croms (preventable waste), because:

1. **Over-classification** — many tasks labeled as needing top-tier models could be handled by Frontier or Balanced models with better prompt structure
2. **Decomposability** — a Mythos-classified step is frequently decomposable into smaller sub-tasks at lower tiers, with the same output quality at a fraction of the cost
3. **Context inefficiency** — top-tier models are most sensitive to context bloat; structural waste multiplies cost fastest at this tier

---

## Related pages

- [Workflow Classification →](/workflow-classification)
- [Croms™ — preventable AI workflow waste →](/croms)
- [Baseline Cost per Workflow →](/baseline-cost-per-workflow)
- [Total Cost of Workflow Ownership →](/total-cost-of-workflow-ownership)