Xiaomi
Xiaomi is a Chinese consumer electronics and internet technology company engaged in the research, development, and sale of smartphones, smart home devices, wearables, and related software services. Its product line spans phones, tablets, televisions, routers, and other categories, centered on an IoT platform-based smart ecosystem. Xiaomi also operates in internet services, artificial intelligence, and new-energy vehicles, with business presence across multiple countries and regions.
Core Facts
Xiaomi officially disclosed that the total training cost of MiMo-V2.6 was only $3.5 million.
The live-stream page simultaneously trains two models, MiMo-V2.6 Pro and MiMo-V2.6 Flash.
Xiaomi has publicly released MiMo-V2.6's model weights, technical report, reinforcement learning (RL) environment, and training code.
Timeline (last 90 days)
Chip vendors (such as Qualcomm and MediaTek) and device vendors (such as Samsung and Xiaomi) contribute code to AOSP, and re-publishing it requires the consent of the contributors, making the coordination cycle lengthy.
Previously, some Huawei and Xiaomi models have already adopted variable aperture technology multiple times.
The portable Bluetooth speaker market is already covered by brands such as JBL, Anker's Soundcore and Xiaomi across price points from entry-level to mid-range.
The mass-production timelines for humanoid robot products such as Tesla Optimus and Xiaomi CyberOne have been repeatedly delayed.
The domestic platforms listed in Pi's official reference table include DeepSeek, Kimi, MiniMax, Alibaba's Tongyi Qianwen, Xiaomi's models, and others.
Xiaomi officially stated that at an equivalent level of intelligence, MiMo-V2.6's cost is one-twentieth to one-sixtieth that of leading overseas models.
The root cause of the tool-call repetition issue in Xiaomi MiMo v2.6 is a reward blind spot in large-scale reinforcement learning—the repetition penalty only takes effect when tool calls exceed 32 times per round.
The updated MiMo model has been open-sourced and went live on the API platform on September 25, with the model name unchanged.
Xiaomi MiMo's fix involved training a lightweight repetition-specific reinforcement learning teacher—approximately 7,000 samples over 12 steps—and then merging it into the main model via MoPD, at a cost of approximately 4% of a full retraining.
The domestic AI models competing in this episode include four: Hunyuan, GLM, DeepSeek, and Xiaomi.
40 more timeline events
All Facts (20)
Xiaomi officially disclosed that the total training cost of MiMo-V2.6 was only $3.5 million.
80%VerifiedThe live-stream page simultaneously trains two models, MiMo-V2.6 Pro and MiMo-V2.6 Flash.
65%UnverifiedMiMo VL Pro总参数量达1T，激活参数42B，支持100万上下文窗口，采用混合专家（MoE）架构
85%UnverifiedMiMo VL TTS在预训练阶段采用了上亿小时的语音数据，支持同一句话内完成语气转折和情感递变
80%Partially VerifiedXiaomi has publicly released MiMo-V2.6's model weights, technical report, reinforcement learning (RL) environment, and training code.
75%UnverifiedThe MiniMax MiMo 3.0 series reconstructs the underlying architecture by sharing and reusing KV cache across layers, reducing prefill computation by approximately 5x and shrinking the KV cache by 4.5x compared to MiMo 2.6 under a 1 million token context.
60%UnverifiedChip vendors (such as Qualcomm and MediaTek) and device vendors (such as Samsung and Xiaomi) contribute code to AOSP, and re-publishing it requires the consent of the contributors, making the coordination cycle lengthy.
50%UnverifiedPreviously, some Huawei and Xiaomi models have already adopted variable aperture technology multiple times.
50%UnverifiedThe portable Bluetooth speaker market is already covered by brands such as JBL, Anker's Soundcore and Xiaomi across price points from entry-level to mid-range.
50%UnverifiedThe mass-production timelines for humanoid robot products such as Tesla Optimus and Xiaomi CyberOne have been repeatedly delayed.
50%UnverifiedThe domestic platforms listed in Pi's official reference table include DeepSeek, Kimi, MiniMax, Alibaba's Tongyi Qianwen, Xiaomi's models, and others.
50%UnverifiedXiaomi officially stated that at an equivalent level of intelligence, MiMo-V2.6's cost is one-twentieth to one-sixtieth that of leading overseas models.
50%UnverifiedThe root cause of the tool-call repetition issue in Xiaomi MiMo v2.6 is a reward blind spot in large-scale reinforcement learning—the repetition penalty only takes effect when tool calls exceed 32 times per round.
50%UnverifiedXiaomi MiMo's fix involved training a lightweight repetition-specific reinforcement learning teacher—approximately 7,000 samples over 12 steps—and then merging it into the main model via MoPD, at a cost of approximately 4% of a full retraining.
50%UnverifiedThe domestic AI models competing in this episode include four: Hunyuan, GLM, DeepSeek, and Xiaomi.
50%UnverifiedXiaomi's imaging collaboration with Leica began with the Xiaomi 12S series in 2022, covering lens color tuning style, lens coatings, and jointly calibrated image processing algorithms.
50%UnverifiedLeica offers Xiaomi two color style options: 'Leica Authentic,' which restores natural tones, and 'Leica Classic,' which delivers richer contrast.
50%UnverifiedBrands such as Samsung and Xiaomi have already established assembly lines in India to avoid some of the tariffs.
50%UnverifiedMajor competitors in India's wearable market include brands such as Noise, boAt, Fire-Boltt, and Xiaomi's Amazfit.
50%UnverifiedXiaomi's MiMo series is an open-source reasoning model released by Xiaomi; MiMo-V2.5 demonstrates strong performance on code and mathematical reasoning tasks and employs reinforcement learning for post-training optimization
50%