aiqbench analysis
Xiaomi and TileRT demonstrate that extreme model-system codesign, combining FP4 quantization with DFlash speculative decoding, can push a trillion-parameter MoE model past 1000 tokens per second on commodity hardware. This achievement highlights inference speed as a critical competitive metric, proving that significant throughput gains are achievable without custom silicon by optimizing the entire stack. For a tech audience, the key takeaway is the viability of selective quantization on MoE experts to alleviate memory bandwidth pressure at scale.
Xiaomi MiMo and TileRT Push a 1-Trillion-Parameter Model Past 1000 Tokens Per Second on Commodity GPUs
Xiaomi's MiMo team, with TileRT, released MiMo-V2.5-Pro-UltraSpeed, a serving mode for the MiMo-V2.5-Pro model. It decodes over 1000 tokens per second on a 1-trillion-parameter model using a single 8-GPU commodity node. The post Xiaomi MiMo and TileRT Push a 1-Trillion-Parameter Model Past 1000 Tokens Per Second on Commodity GPUs appeared first on MarkTechPost.
Inference speed is becoming a competitive metric for large language models. Xiaomi’s MiMo team just released MiMo-V2.5-Pro-UltraSpeed, built in collaboration with the TileRT systems group. It decodes faster than 1000 tokens per second on a 1-trillion-parameter model. Xiaomi team describes this as a first at trillion-parameter scale. Demos show generation peaks near 1200 tokens per second. The notable part is the hardware: it runs on commodity GPUs, not custom silicon. What is MiMo-V2.5-Pro-UltraSpeed UltraSpeed is a high-speed serving mode for the existing MiMo-V2.5-Pro model. The base model uses a Mixture-of-Experts (MoE) architecture at trillion-parameter scale. UltraSpeed targets generation speed rather than model capability. It changes how fast the model produces output tokens.…
Excerpt shown under the publisher's RSS terms · 112 words · full text and figures are on the publisher's site
Related coverage
Open models recap: more on Kimi K3, Qwen 3.8, Xi's WAIC speech, distillation, the open-closed gap, and what's next
Interconnects · Jul 22, 2026
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026
[AINews] Xiaomi MiMo-V2.6-Pro 1T-A42B: the new top Open Weights model, trained for $3M
Latent Space · Sep 22, 2026
GEM Training: How Meta Doubled the Efficiency of Its LLM-Scale Ads Foundation Model
Meta Engineering · Aug 3, 2026