[{"data":1,"prerenderedAt":30},["ShallowReactive",2],{"nr-en-xiaomi-mimo-v2-6-pro-guenstig-spitzenmodell":3},{"slug":4,"title":5,"dek":6,"date":7,"time":8,"publishedAt":9,"updated":10,"updatedAt":10,"dateFmt":11,"updatedFmt":10,"kind":12,"tier":13,"author":14,"authorName":15,"topics":16,"tracker":22,"trackerLabel":23,"headlineStat":24,"image":25,"ogImage":26,"imageAlt":5,"csv":10,"minutes":27,"words":28,"html":29},"xiaomi-mimo-v2-6-pro-guenstig-spitzenmodell","Xiaomi Launches Budget-Friendly Top-Tier KI Model MiMo-V2.6-Pro","Xiaomi's new model leads rankings of open-source AI systems while undercutting competitors drastically on price. The success is built on massive reinforcement learning – yet Anthropic accuses the company of misusing Claude data.","2026-09-22","14:34","2026-09-22T14:34:00+02:00","","September 22, 2026","news","standard","ideal-syka","Ideal Syka",[17,18,19,20,21],"AI Models","Reinforcement Learning","Xiaomi","Open Source AI","Market Dynamics","\u002Fki-iq","AI Intelligence Ranking","46 points on Intelligence Index – rank 1 among open-source AI models","\u002Fnewsroom\u002Fimg\u002Fxiaomi-mimo-v2-6-pro-guenstig-spitzenmodell.webp","\u002Fog-nr\u002Fxiaomi-mimo-v2-6-pro-guenstig-spitzenmodell.en.png",3,535,"\u003Cp>Xiaomi has released \u003Cstrong>MiMo-V2.6-Pro\u003C\u002Fstrong>, an AI model that leads rankings of open-source systems while upending the industry's price expectations. According to the company, the top-tier model achieves \u003Cstrong>46 points on the Intelligence Index\u003C\u002Fstrong> from analyst firm Artificial Analysis, making it the strongest freely available model to date – ahead of competitors like Qwen. The decisive difference lies in pricing: a test task costs just around \u003Cstrong>$0.13 USD\u003C\u002Fstrong>, a fraction of what comparably powerful models charge.\u003C\u002Fp>\n\u003Ch2>Key Facts\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Cstrong>MiMo-V2.6-Pro\u003C\u002Fstrong> scores 46 points on the Intelligence Index, leading open-source AI rankings\u003C\u002Fli>\n\u003Cli>Cost per token: \u003Cstrong>$0.435 per million input tokens, $0.87 per million output tokens\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>Reinforcement learning training took less than \u003Cstrong>6 days\u003C\u002Fstrong> and cost \u003Cstrong>$2.62 million\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>Xiaomi publicly releases RL tools, training framework, and \u003Cstrong>7,000 ready-made training tasks\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Anthropic accuses Xiaomi\u003C\u002Fstrong> of unauthorized use of Claude training data\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>The Price-Performance Breakthrough Through Reinforcement Learning\u003C\u002Fh2>\n\u003Cp>Xiaomi attributes the performance leap to significantly expanded \u003Cstrong>reinforcement learning\u003C\u002Fstrong> – a training method where models learn through trial, feedback, and reward. The company scaled this phase along three axes: larger data volumes per training step, more diverse task environments, and greater computational power for evaluating solutions.\u003C\u002Fp>\n\u003Cp>The results are impressive: on the \u003Cstrong>DeepSWE\u003C\u002Fstrong> programming benchmark, the Pro model's score climbed from 58.4 to 72.6, while the smaller Flash model jumped from 48.8 to 65.7. Training for Pro took less than six days and cost around \u003Cstrong>$2.62 million\u003C\u002Fstrong>; the Flash model was developed for \u003Cstrong>$0.85 million\u003C\u002Fstrong>.\u003C\u002Fp>\n\u003Cp>Technically, MiMo-V2.6-Pro is a \u003Cstrong>mixture-of-experts model\u003C\u002Fstrong> with 1.02 trillion parameters, of which only 42 billion are active per query. This keeps the system efficient despite its scale.\u003C\u002Fp>\n\u003Ch2>Openness With Questions\u003C\u002Fh2>\n\u003Cp>What stands out is that Xiaomi openly releases its RL tools: the technical report, complete training framework, a smaller model for further training, and roughly \u003Cstrong>7,000 ready-made training tasks\u003C\u002Fstrong> with automated checkers. These cover software development, cybersecurity, office work, web design, and approximately 1,000 music composition tasks.\u003C\u002Fp>\n\u003Cp>Yet this demonstrative openness contrasts sharply with allegations that \u003Cstrong>Anthropic raised just two weeks earlier\u003C\u002Fstrong>: the Claude developer accuses Xiaomi of unauthorized access to training data about Claude to train its own MiMo model series. Xiaomi has not yet responded to these allegations.\u003C\u002Fp>\n\u003Cdiv class=\"tbl-scroll\">\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Aspect\u003C\u002Fth>\n\u003Cth>MiMo-V2.6-Pro\u003C\u002Fth>\n\u003Cth>MiMo-V2.6-Flash\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>Intelligence Index\u003C\u002Ftd>\n\u003Ctd>46 points\u003C\u002Ftd>\n\u003Ctd>–\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Parameters\u003C\u002Ftd>\n\u003Ctd>1.02 trillion\u003C\u002Ftd>\n\u003Ctd>–\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Active Parameters\u003C\u002Ftd>\n\u003Ctd>42 billion\u003C\u002Ftd>\n\u003Ctd>–\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>RL Training Cost\u003C\u002Ftd>\n\u003Ctd>$2.62M\u003C\u002Ftd>\n\u003Ctd>$0.85M\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Training Duration\u003C\u002Ftd>\n\u003Ctd>< 6 days\u003C\u002Ftd>\n\u003Ctd>< 6 days\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2>What This Means for European Companies\u003C\u002Fh2>\n\u003Cp>The announcement signals a shift in AI market dynamics: established hardware manufacturers like Xiaomi are entering a field long dominated by OpenAI, Anthropic, and Deepseek with their own top-tier models. The aggressive price point could pressure European AI providers and further intensify cost pressures for companies using AI models. Simultaneously, the accusation against Xiaomi raises questions about data security and protection of proprietary training methods – issues that could become relevant for German firms handling sensitive data if they train their models with open systems.\u003C\u002Fp>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fthe-decoder.de\u002Fxiaomis-mimo-v2-6-wird-staerkstes-offenes-ki-modell-und-bleibt-dabei-guenstig\u002F\">The Decoder (DE)\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cem>Editorially owned by \u003Ca href=\"\u002Fen\u002Fautor\u002Fideal-syka\">Ideal Syka\u003C\u002Fa>. Sources and method: \u003Ca href=\"\u002Fen\u002Fredaktion\">Newsroom & method\u003C\u002Fa>. Tips and corrections: \u003Ca href=\"mailto:ai@i6eal.de\">ai@i6eal.de\u003C\u002Fa>.\u003C\u002Fem>\u003C\u002Fp>\n",1790080883104]