Mimo 2.5 Pro - 40t/s on 8x Nvidia Spark/GB10 cluster
The Mimo 2.5 Pro large language model has been benchmarked on an 8x Nvidia GB10 cluster, achieving impressive throughput speeds. Under single-user conditions, it reached 40 tokens/second with a 1k context, scaling up to 17 tokens/second with a 250k context. With parallel processing, the model demonstrated even higher performance, hitting 83 tokens/second with four parallel requests. AI
IMPACT Demonstrates high throughput for large context windows on specialized hardware, potentially influencing local LLM deployment strategies.