Mythos Preview Leads in Long Context

scaling01 · x · 2026-07-17

This post relays a model comparison result: **Mythos Preview** outperforms **GPT-5.6-Sol** and **Mythos 5** in the first approximately **30 million token** long-context phase. The author adds that Mythos Preview's best attempt speed is also slightly faster than that of GPT-5.6-Sol. A cited comment provides another comparison: in UK AISI's long-range cyber attack and defense tests, **open-models** lag behind frontier models by about **7 months**. Notably, **GLM-5.2** is closing in on **Opus 4.5** on the `The Last Ones` cyber range, and matches **Opus 4.6** on shorter, narrower tasks.

Original post →

More from Models

Models channel →