AIK2 Horizon’s 3.7B and 7B models reportedly drew much larger size guesses at ICML
IFM says its July 2026 preview drew average guesses of 89.9 billion and 62.6 billion parameters, respectively, after attendees watched real-work and coding tasks.
TLDR
IFM says it previewed K2-Horizon-3.7B and 7B at ICML in July 2026, asking attendees to guess their sizes while watching real-work and coding tasks. According to IFM, the average guesses were 89.9 billion and 62.6 billion parameters, respectively—well above the models’ actual counts. IFM argues that intelligence density now matters more than parameter count and says K2 Horizon is built around capable, smaller models, lower cost and higher speed. Its announcement links to both models on Hugging Face.
K2 Horizon’s 3.7B and 7B models reportedly drew much larger size guesses at ICML
IFM says its July 2026 preview drew average guesses of 89.9 billion and 62.6 billion parameters, respectively, after attendees watched real-work and coding tasks.
