Sakana AIContinuous Thought Machines
ctm-maze-large
This CTM checkpoint was trained on 39 by 39 mazes. It outputs movement sequences rather than conversation or prose.
Research release
The study examines navigation beyond the training setup
The card reports navigation on larger 99×99 mazes by extending the model's processing. This is a result for the reported maze experiment, not a guarantee for arbitrary planning problems.ctm-maze-large official model card
Working with this release
Practical guidance. Check that every predicted move is legal and that the path reaches the goal. Compare different maze sizes without conflating more computation with a larger trained model.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | SakanaAI/ctm-maze-large |
| Actions | Left, right, up, down and padding.ctm-maze-large official model card |
| Training setup | 39×39 mazes and a 100-step training limit; no positional embeddings.ctm-maze-large official model card |
| Terms | Apache 2.0; experimental prototype, not intended for commercial or critical deployment.ctm-maze-large official model card |
Related models
- ctm-imagenet
Sakana AI · Continuous Thought Machines
An iterative neural model trained for ImageNet image classification.
Launch date not confirmedResearch releaseDownloadable