TechnologyTuesday, October 20, 20261 min read

Inception launches Mercury 3, a diffusion-based LLM it claims is "comparable to Luna and Haiku models" but 4x to 8x faster, generating ~1,000 tokens per second (Jason Hiner/The Deep View)

By Techmeme · Published by Everything Blog

In short

Inception introduces Mercury 3, a diffusion-based language model that is claimed to be comparable to Luna and Haiku models but significantly faster, generating approximately 1,000 tokens per second. This model is positioned as a powerful tool for businesses and developers seeking to enhance their AI capabilities without compromising speed or performance.

Key points

It is designed for businesses and developers seeking enhanced AI capabilities.: It is designed for businesses and developers seeking enhanced AI capabilities.
Inception claims it is 4x to 8x faster than its predecessor, Mercury 2.: Inception claims it is 4x to 8x faster than its predecessor, Mercury 2.

Sources checked by our analysis pipeline: 1

Connor Hart / Wall Street Journal: |

Chance Miller / 9to5Mac: |

MiniMax: |

Dominic-Madori Davis / TechCrunch: |

Bloomberg: |

Cloudflare: |

Todd Bishop / GeekWire: |

Financial Times: |

Raphael Satter / Reuters: |

Bloomberg: |

ProPublica: |

Sarah Perez / TechCrunch: |

Ellen DiMauro / Semafor: |

Todd Bishop / GeekWire: |

Dan Mangan / CNBC: |

Brian Krebs / Krebs on Security: |

Engineering at Meta: |

Reuters: |

Wall Street Journal: |

John Koetsier / Forbes: |

Kai Nicol-Schwarz / CNBC: |

The Information: |

Bloomberg: |

Stephanie Palazzolo / The Information: |

Bloomberg: |

Jason Hiner / The Deep View: |

Matthias Bastian / The Decoder: |

Alex Konrad / Upstarts Media: |

Paige Smith / Bloomberg: |

This is a Techmeme archive page. It shows how the site appeared at 9:45 PM ET, October 19, 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.

Maddy Varner / Wired: |

Caroline Haskins / Wired: |

Zack Abrams / The Block: |

Original source: www.techmeme.com

How HP Is Building an AI-Ready CDP with Databricks CustomerLake
HP has modernized its marketing data foundation with Databricks CustomerLake and Genie, accelerating audience creation and reducing costs by up to 40%, expanding available marketing data by 90%, and changing how teams work together by providing faster access to data for planning, targeting, and measurement. This approach can reduce costs, expand available customer context, and accelerate activation, materializing in HP's early experience with CustomerLake.
4 min read
Go will reject certificate chains containing eight distrusted SPKIs
Go has decided to reject certificate chains containing eight distrusted Subject Public Key Infrastructures (SPKIs) as part of a security effort to address CVE-2026-111943, a critical vulnerability in the Go library. This change has been made in two branches: release-branch.go1.26 and release-branch.go1.27, ensuring that the security concerns are mitigated. The change has been acknowledged by several contributors, including @thatnealpatel, and the necessary cherry-pick CLs have been created for the affected release…
1 min read
‘Open your doors for AI’: German chancellor urges businesses to embrace AI as Nvidia announces new investments
Nvidia is increasing its investments in Europe's artificial intelligence sector, unveiling plans to back four German startups as the country pushes to strengthen its position in industrial AI. chipmaker's nearly $10 billion commitment to invest in companies, universities and research institutions across Europe, as announced at the Invest in Germany Summit in Berlin on Monday. Nvidia CEO Jensen Huang echoed Chancellor Friedrich Merz's remarks while highlighting Germany's expertise in manufacturing, mathematics and…
2 min read

Inception launches Mercury 3, a diffusion-based LLM it claims is "comparable to Luna and Haiku models" but 4x to 8x faster, generating ~1,000 tokens per second (Jason Hiner/The Deep View)

In short

Key points

More in Technology