Artificial Intelligence1 min read

TML releases Inkling 2 Mini, an open-weights 26.4B-parameter transformer that loops middle layers three times per token for better reasoning with fewer tokens (Thinking Machines Lab)

By · Published by Everything Blog

In short

Thinking Machines Lab has made the model available to the broader community with open weights. The system achieves enhanced reasoning efficiency by cycling through its middle transformer layers three times for each individual token. This architectural choice demonstrates how recursive layer processing can strengthen computational logic within a 26.4 billion parameter footprint.

Key points

  • Recursive looping of middle layers allows the model to reason effectively with fewer toke…: Recursive looping of middle layers allows the model to reason effectively with fewer tokens.
  • Open weights provide the broader AI community direct access to this experimental architec…: Open weights provide the broader AI community direct access to this experimental architecture.
  • Improving middle-layer utilization offers an alternative to expanding raw parameter count…: Improving middle-layer utilization offers an alternative to expanding raw parameter counts for complex reasoning.

Sources checked by our analysis pipeline: 1

New York Times: |

Bloomberg: |

Kif Leswing / CNBC: |

Wired: |

Joseph Howlett / Scientific American: |

Ashley Capoot / CNBC: |

William Sandlund / Financial Times: |

Reuters: |

CNBC: |

Katie Nicora / CBS News: |

Katie Paul / Reuters: |

Jennifer Maas / Variety: |

Grace Huckins / MIT Technology Review: |

Julianna Bragg / Axios: |

Ben Schoon / 9to5Google: |

Barry Schwartz / Search Engine Roundtable: |

Harper Harwell / Birmingham Business Journal: |

Axios: |

Courtney Rozen / Reuters: |

Olivia Carville / Bloomberg: |

Bloomberg: |

Thinking Machines Lab: |

Bloomberg: |

This is a Techmeme archive page. It shows how the site appeared at 4:15 PM ET, November 20, 2026.

The most current version of the site as always is available at our home page. To view an earlier snapshot click here and then modify the date indicated.

Kirsten Korosec / TechCrunch: |

Yoolim Lee / Bloomberg: |

Tim Genewein / DeepMind Institute: |

Mark Gurman / Bloomberg: |

Sarah Perez / TechCrunch: |

Artificial Intelligence