# SakanaAI/continuous-thought-machines

Continuous Thought Machines, because thought takes time and reasoning is a process.

Repository: https://github.com/SakanaAI/continuous-thought-machines
Canonical: https://ross.abutalabs.com/products/continuous-thought-machines
Homepage: https://pub.sakana.ai/ctm/
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2025-12-29T03:52:23+00:00

## Health v2 (maintenance only)
Score: 46/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 59, release rhythm 35, longevity 34
- inputs: {"age_days": 480, "days_push": 247, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2019, forks 307 (observed 2026-08-28T04:06:06.252538+00:00)

## What it is
The Continuous Thought Machine (CTM) is a neural network architecture from Sakana AI that uses neuron-level temporal dynamics and neural synchronization as its core representation mechanism. The repository provides PyTorch implementations, training code, and analysis tools for tasks like ImageNet classification, maze solving, sorting, parity, question-answering, and reinforcement learning.

## Use cases
- implement continuous thought machine models
- train a neural network that reasons over internal time steps
- reproduce CTM research results on ImageNet or mazes
- explore neuron-level temporal dynamics in deep learning
- build biologically inspired reasoning architectures
- run maze-solving neural network demos
- study neural synchronization as a latent representation

## When to choose
- you want to experiment with or extend the CTM architecture
- you need reproducible code for the CTM research paper
- you're researching temporal neural dynamics or biologically plausible AI
- you want a model that unfolds reasoning over an internal time axis

## When to avoid
- you need a production-ready inference model for standard tasks
- you want a lightweight off-the-shelf classifier rather than a research codebase
- your project requires minimal compute - CTMs are research-heavy and resource intensive

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning
- domain: artificial-intelligence, deep-learning, machine-learning
- platform: python
- tags: neural-dynamics, research-code, temporal-processing, neural-synchronization, biologically-inspired, pytorch

## Member repositories
- SakanaAI/continuous-thought-machines (main) score 46

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:06.252538+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 2026-08-30T03:00:25.923011+00:00, confidence not recorded.
  - readme: https://github.com/SakanaAI/continuous-thought-machines (fetched 2026-08-28T04:06:06.252538+00:00, sha da3f12890660)
  - homepage: https://pub.sakana.ai/ctm/ (fetched 2026-08-29T10:40:18.241183+00:00, sha 44eda85cea5f)
- Data as of 2026-08-30T08:39:29.467469+00:00.