~/AI FOR SCIEN/xiaomi-open-sources-mimo-v2-6-multimodal-ai-models-for-phd-level

Xiaomi Open-Sources MiMo-V2.6 Multimodal AI Models for PhD-Level Scientific Research

Xiaomi officially released and open-sourced its MiMo-V2.6 multimodal AI series, featuring Pro and Flash models designed to advance recursive self-improvement (RSI). The MiMo-V2.6-Pro model demonstrated PhD-level assistance in material design for PFAS pollutant adsorption and completed a 6,000-line formal mathematical proof in Lean 4 without domain-specific post-training. This release highlights how general-purpose reasoning and multi-agent collaboration can significantly accelerate scientific discovery, reducing material design cycles from a month down to 2–3 days. Furthermore, offering high-level multimodal capabilities at a fraction of the cost of foreign competitors helps democratize AI for Science (AI4S). In mathematical reasoning, MiMo-V2.6-Pro formally proved the Li–Yorke theorem ("Period Three Implies Chaos") in Lean 4 with over 6,000 lines of code passing kernel verification. API pricing remains accessible, with Flash priced at 1 RMB input / 2 RMB output per million tokens and Pro at 3 RMB input / 6 RMB output, alongside a 99% cache discount.

## BACKGROUND

Metal-Organic Frameworks (MOFs) are porous crystalline materials composed of metal ions coordinated to organic linkers, widely researched for gas storage and environmental remediation. Lean 4 is an open-source interactive theorem prover and functional programming language used to rigorously verify mathematical proofs.

## REFERENCES

## KEYWORDS

#AI for Science#LLM#Open Source AI#Multimodal AI#Formal Verification

$ subscribe --daily

Xiaomi Open-Sources MiMo-V2.6 Multimodal AI Models for PhD-Level Scientific Research | Daily News