Achieved top score on METR time horizon benchmark, handling 90-minute software tasks

Extended context window to 1 million tokens, placing it among models with longest current context capabilities

Achieved state-of-the-art status on most benchmarks according to preliminary evaluations