CLAIM-2026-09-08-02

Each K2 Horizon model was pre-trained on approximately 20 trillion tokens, using in IFM's words "carefully constructed and documented mixtures".

Why it matters

It establishes the scale of the original training run, which is what the downstream-modifier threshold would be measured against if the compute figure were published.

Sources

• instituteof2026introduci
• cellcog2026k
• marktechpos2026ifm