The End of the Context Window: 116 Million Tokens, One Desktop, 84.5%

Mneme + Atlas posted 84.5% on full BEAM-10M (208,696 turns, ~116 million tokens) with multi-session reasoning at 95%, on one NVIDIA DGX Spark desktop GPU using a 35B open model and a 32K context window.

Contact: enterprise@native.business · Dubai, UAE