Llama-3.1 Nemotron Ultra 253B vs MiMo-V2-Flash

Llama-3.1 Nemotron Ultra 253B

NVIDIA

Llama-3.1-Nemotron-Ultra-253B-v1 is a 253-billion-parameter model from NVIDIA, derived from Meta's Llama 3.1 405B using neural architecture search (NAS) compression combined with NVIDIA's Nemotron post-training pipeline, which recovers and exceeds the base model's capability after structural compression. Released in April 2025, it supports toggling between a standard instruction mode and an extended reasoning mode via system prompt, allowing the same model to handle both rapid responses and deliberate chain-of-thought tasks. It is the flagship of the Nemotron family, available open-weight on HuggingFace and through NVIDIA NIM for enterprise inference.

Xiaomi

MiMo-V2-Flash, released by Xiaomi on December 16, 2025, is a Mixture-of-Experts large language model with 309 billion total parameters and 15 billion active parameters per inference, designed for high-speed reasoning and agentic workflows. It features a 256K token context window, processes up to 150 tokens per second, and was trained on 27 trillion tokens. MiMo-V2-Flash targets open-source deployments requiring fast, capable coding and reasoning with an efficient inference footprint, under an MIT license.