Saturday, November 21, 2026← Front page
Also this week

K2-Horizon open models

K2 Horizon is among the projects that opened up their training runs, according to a Latent Space newsletter roundup.1 The only source here covers K2 Horizon briefly, alongside Marin's release of 57 intermediate checkpoints from its 535B-A23B run, and offers no reactions or disputes specific to K2 Horizon. inference21

The K2 report compares four model sizes that were all trained on the same 22T tokens.3 It also describes a decoding adapter that the team says achieves a 2.2× speedup.4

What each sentence rests on
  1. Marin and K2 Horizon opened up their training runs. — Latent Space, [AINews] GPT-6.2 Astra, the #2 Frontier Model is better and 20%-600% cheaper, but has mixed reviews; 6.2 Sol likely main workhorse
  2. Marin released 57 intermediate Mochi checkpoints from its 535B-A23B run. — Latent Space, [AINews] GPT-6.2 Astra, the #2 Frontier Model is better and 20%-600% cheaper, but has mixed reviews; 6.2 Sol likely main workhorse
  3. The K2 report compares four model sizes trained on the same 22T tokens — Latent Space, [AINews] GPT-6.2 Astra, the #2 Frontier Model is better and 20%-600% cheaper, but has mixed reviews; 6.2 Sol likely main workhorse
  4. describes a decoding adapter that the team says reaches 2.2× speedup — Latent Space, [AINews] GPT-6.2 Astra, the #2 Frontier Model is better and 20%-600% cheaper, but has mixed reviews; 6.2 Sol likely main workhorse

Every sentence was checked word for word against quotes from the articles it cites. Each footnote shows its quote, unless the page already quotes as much of that article as we allow; then it links the article instead. A model wrote and arranged the sentences, and a second read checked each against its quotes (written 2026-11-20).

The record: who reported it, checked facts, all quoted views, timeline →