@zephyr_z9: This is super big I think this is the first useful speculative decoding method deployed on a big quasi frontier model M…
Summary
Xiaomi MiMo releases MiMo-V2.5-Pro-UltraSpeed, achieving over 1,000 tokens per second on a 1 trillion parameter model using speculative decoding, the first practical deployment of such speed at scale.
View Cached Full Text
Cached at: 06/09/26, 10:45 AM
This is super big I think this is the first useful speculative decoding method deployed on a big quasi frontier model Massive unlock @fi56622380 https://t.co/augiaFLDOK
Xiaomi MiMo (@XiaomiMiMo): 🚀 1,000+ TOKENS/S ON A 1T MODEL! 🚀
We are thrilled to release Xiaomi MiMo-V2.5-Pro-UltraSpeed in collaboration with @TileRT_AI , breaking the 1,000 tokens/s output speed on a 1 Trillion parameter model for the FIRST TIME!
Not wafer-scale integration like Cerebras. Not pure
Similar Articles
China's Xiaomi MiMo Is Now 15X Faster Than ChatGPT and Claude (4 minute read)
Xiaomi achieved over 1,000 tokens per second inference on its trillion-parameter MiMo-V2.5-Pro-UltraSpeed model using commodity 8-GPU nodes via FP4 quantization and DFlash speculative decoding, outpacing GPT-5.5 and Claude Opus by over 10x.
XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash
XiaomiMiMo releases MiMo-V2.5-Pro-FP4-DFlash, an FP4-quantized MoE model with block-diffusion speculative decoding to reduce memory and bandwidth for trillion-parameter inference.
Xiaomi just claimed 1,000+ tps on a 1T model using a standard 8-GPU server
Xiaomi released MiMo-V2.5-Pro-UltraSpeed in collaboration with TileRT, achieving over 1000 tokens/s decode speed on a 1-trillion-parameter model, enabling real-time AI interaction and accelerating coding agents and reasoning tasks.
@charles_irl: Speculation Is All You Need. In this blog post, we announce the co-release (w/ Z Lab) of six more state-of-the-art DFla…
Modal and Z Lab release six new DFlash speculative decoding draft models for Qwen 3.x, achieving over 1000 tokens per second on a B200 and arguing that speculative decoding is the most impactful inference optimization.
@rohanpaul_ai: New @cohere paper finds that speculative decoding works well on fine-grained MoE models even at batch size 1, where pri…
A new Cohere paper shows speculative decoding works well on fine-grained MoE models even at batch size 1, since adjacent tokens share 20%–45% of experts (vs 4%–12% expected), making verification far cheaper than prior work predicted across 10 MoE models tested on H100s.