# kyegomez/OpenMythos

A theoretical reconstruction of the Claude Mythos architecture, built from first principles using the available research literature.

Repository: https://github.com/kyegomez/OpenMythos
Canonical: https://ross.abutalabs.com/products/openmythos
Homepage: https://discord.gg/EamjgSaEQf
Language: Python
License: MIT
License Family: permissive
Topics: ai, attention, claude, claude-ai, claude-code, gpt-5, looped-transformers, ml, sonnet, gpt-7, anthropic, claude-code-plugin, claude-mythos, claude-sonnet, deepmind, jax, pytorch, torch
Last push: 2026-05-23T11:58:48+00:00

## Health v2 (maintenance only)
Score: 51/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 83, release rhythm 35, longevity 9
- inputs: {"age_days": 137, "days_push": 102, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 14805, forks 3293 (observed 2026-08-28T04:11:07.726298+00:00)

## What it is
OpenMythos is an open-source, theoretical PyTorch implementation of a Recurrent-Depth Transformer (RDT) architecture inspired by speculation about Anthropic's Claude models. It features a Prelude/looped Recurrent Block/Coda structure with switchable MLA or GQA attention and sparse MoE feed-forward layers.

## Use cases
- implement a looped recurrent-depth transformer from scratch
- experiment with compute-adaptive depth-variable reasoning in transformers
- compare MLA vs GQA attention implementations
- train a sparse mixture-of-experts transformer with routed and shared experts
- study theoretical reconstructions of proprietary LLM architectures
- prototype adaptive inference loops in PyTorch

## When to choose
- you want to experiment with recurrent-depth or looped transformer architectures in PyTorch
- you need a configurable MoE transformer with switchable attention backends
- you are researching compute-adaptive reasoning and want a hackable codebase

## When to avoid
- you need a production-ready, officially supported Claude model or API
- you require guaranteed fidelity to Anthropic's proprietary architecture
- you need a battle-tested framework for training large models at scale

## Facets
- artifact type: library
- maturity: experimental
- function: machine-learning, deep-learning, llm-training, llm-inference
- domain: deep-learning, large-language-models, machine-learning
- platform: python
- tags: transformer, recurrent-depth, mixture-of-experts, pytorch, jax, attention, research-implementation, looped-transformers, gpu

## Member repositories
- kyegomez/OpenMythos (main) score 51

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:11:07.726298+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:07:08.998295+00:00, confidence not recorded.
  - readme: https://github.com/kyegomez/OpenMythos (fetched 2026-08-28T04:11:07.726298+00:00, sha 154dfdd50ead)
  - homepage: https://discord.gg/EamjgSaEQf (fetched 2026-08-29T08:05:06.581231+00:00, sha b6a9fbc44722)
- Data as of 2026-08-30T08:39:29.467469+00:00.