Xiaomi open-sources MiMo-V2.6-Pro, scoring 46 on AI Intelligence Index as top open-weight model
On September 22, Xiaomi released and open-sourced its MiMo-V2.6 series of large AI models, including Pro and Flash variants. The MiMo-V2.6-Pro scored 46 on the Artificial Analysis Intelligence Index, surpassing Kimi K3 and Qwen3.8 Max to become the highest-ranked open-source model, though still behind closed-source leaders Claude Fable 5.1 and GPT-6 Astra. API pricing remains unchanged from the V2.5 series. Training costs revealed on September 17 showed $890,000 for Pro and $397,000 for Flash. Xiaomi open-sourced the technical report, training environment, and RL code.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary awaiting refresh
Summary awaiting refresh
Cross-source coverage
Reporting timeline
Xiaomi Releases and Open-Sources MiMo-V2.6 Series Large Models, Keeps V2.5 API Pricing
On September 22, Xiaomi released and open-sourced its Xiaomi MiMo-V2.6 series of large models, described as a key step in exploring recursive self-improvement (RSI) by scaling reinforcement learning (RL) compute on verifiable complex tasks. The series includes two native full-modal models: MiMo-V2.6-Pro and MiMo-V2.6-Flash. According to Xiaomi, MiMo-V2.6-Pro scored 46 on the Artificial Analysis Intelligence Index, surpassing Kimi K3 and Qwen3.8 Max to become the strongest open-source model, though still behind closed-source leaders Claude Fable 5.1 and GPT-6 Astra. API pricing remains unchanged from the V2.5 series. On September 17, MiMo team lead Luo Fuli shared real-time training pages on X, revealing that MiMo-V2.6-Pro training cost $890,000 over 1 day 19 hours (over $20,000/hour), and MiMo-V2.6-Flash cost $397,000 over 1 day 14 hours (over $10,000/hour), totaling about $31,000 per hour. Xiaomi expanded RL compute across three dimensions, supporting 1M context length training with 2.7–3.7B tokens per step, and built a multi-task training system covering Code, General, Visual, and Cyber directions. Xiaomi has open-sourced the technical report, training environment, and RL code.
Read sourceXiaomi Releases MiMo-V2.6 Open-Source AI Model, Claims It Is the Strongest Open-Source Model
On September 22, Xiaomi officially released and open-sourced its new Xiaomi MiMo-V2.6 series of AI models, including Pro and Flash variants. The company claims that due to expanded reinforcement learning compute, the MiMo-V2.6-Pro model scored 46 on the Artificial Analysis Intelligence Index (AA Comprehensive Intelligence Index), surpassing Kimi K3 and Qwen3.8 Max to become the strongest open-source model currently available. However, it still trails behind leading closed-source models such as Claude Fable5.1 and GPT-6 Astra. Xiaomi stated that the MiMo-V2.6 series maintains the same API pricing as the V2.5 series, and that MiMo-V2.6-Pro sets a new cost-performance record for domestic models, with prices ranging from 1/20 to 1/60 of comparable overseas models at the same intelligence level.
Read sourceXiaomi Releases MiMo-V2.6 Model, Claims It Is the Strongest Open-Source Model
On September 22, Jin10 Data reported that Xiaomi officially released and open-sourced the new Xiaomi MiMo-V2.6 series early this morning. The team stated that the MiMo-V2.6 series includes two native full-modal models: Pro and Flash. Thanks to expanded RL computing power, MiMo-V2.6-Pro scored 46 points on the Artificial Analysis Intelligence Index (AA Comprehensive Intelligence Index), surpassing Kimi K3 and Qwen3.8 Max to become the current strongest open-source model. However, it still lags behind the strongest closed-source models, Claude Fable5.1 and GPT-6 Astra. The team noted that the MiMo-V2.6 series maintains the same API pricing as the V2.5 series, with MiMo-V2.6-Pro setting a new cost-performance record for domestic models: at the same intelligence level, its price is only 1/20 to 1/60 of overseas models.
Show 5 older updatesHide older updates
Xiaomi releases MiMo-V2.6-Pro and Flash open-source AI models, ranking 6th on intelligence index
Xiaomi has released its MiMo-V2.6-Pro and Flash open-source models, according to a post on X. The MiMo-V2.6-Pro model has achieved the 6th spot on the Artificial Analysis Intelligence Index, making it a top-performing open-weight model available. The post claims the model performs on par with Claude Opus 5 and GPT-5.6 Sol. The announcement invites the open-source community to begin testing the models.
Read sourceXiaomi MiMo Releases MiMo-V2.6 Pro and Flash Open-Source Multimodal Models
Xiaomi MiMo announced the release of MiMo-V2.6 Pro and Flash, two full-modal models trained through large-scale reinforcement learning. According to the announcement, the Pro version performs comparably to Claude Opus 5 and GPT-5.6 Sol on most agent benchmarks. It scored 46 on the Artificial Analysis Intelligence Index, the highest among open-source models. The models' capabilities cover coding, computer use, 3D reasoning, and creative tasks. Xiaomi MiMo has made the model weights, technical report, and training code publicly available, along with benchmark comparison data against Claude Opus 5 and GPT-5.6 Sol.
Read sourceXiaomi's MiMo-V2.6-Pro tops open weights AI models with Intelligence Index of 46 at $0.13 per task
Xiaomi has released MiMo-V2.6-Pro, an open weights AI model that has achieved the top position on the Artificial Analysis Intelligence Index with a score of 46. The model costs $0.13 per Intelligence Index task, placing it on the Intelligence vs. Cost per Task Pareto frontier for efficiency. Compared to its predecessor, MiMo-V2.5-Pro (Intelligence Index: 26), the new model shows major intelligence advances while maintaining the same pricing: $0.435 per 1M input tokens (with a 99% cache-hit discount) and $0.87 per 1M output tokens. MiMo-V2.6-Pro is a Mixture-of-Experts (MoE) model with 1.02 trillion total parameters and 42 billion active parameters. The source notes that further analysis of the model is forthcoming and provides a link to a full benchmarking breakdown.
Read sourceXiaomi Launches MiMo-V2.6-Pro AI Model with $0.13 Task Cost, Bringing Frontier Intelligence to $0.1 Range
On September 22, Xiaomi released and open-sourced its next-generation MiMo large language models: the flagship multimodal model Xiaomi MiMo-V2.6-Pro and the efficient inference model Xiaomi MiMo-V2.6-Flash, along with an ultra-speed mode and a desktop client. According to the global AI evaluation platform Artificial Analysis, the MiMo-V2.6-Pro achieved a measured single-task cost of $0.13, bringing frontier AI intelligence into the $0.1 cost range for the first time. The model scored 46 on the Artificial Analysis comprehensive intelligence index, making it one of the strongest open-source models. While maintaining the same price as the previous generation, the model's intelligence reportedly doubled. Notably, MiMo-V2.6-Pro began participating in real scientific research as a 'Co-Scientist,' assisting Xiaomi's advanced materials research team in designing metal-organic framework (MOF) materials for adsorbing PFAS pollutants, improving research efficiency by approximately 10 times. Peking University Professor Dou Jinhu commented that the model's research ability reached the level of a trained doctoral researcher. The model also helped formalize the Li–Yorke theorem in Lean 4. Xiaomi open-sourced the model weights, RL training techniques, and related resources.
Read sourceXiaomi Releases and Open-Sources MiMo-V2.6 AI Model Series, Keeps V2.5 API Pricing
On September 22, Xiaomi released and open-sourced its Xiaomi MiMo-V2.6 series of large AI models, described as a key step in exploring recursive self-improvement (RSI) by scaling reinforcement learning (RL) compute on verifiable complex tasks. The series includes two native multimodal models: MiMo-V2.6-Pro and MiMo-V2.6-Flash. According to the company, MiMo-V2.6-Pro achieved top rankings in the Artificial Analysis Intelligence Index. The API pricing remains unchanged from the V2.5 series. On September 17, Xiaomi's MiMo model team leader Luo Fuli shared real-time training pages on X, revealing that MiMo-V2.6-Pro training cost $890,000 over 1 day 19 hours (over $20,000/hour), and MiMo-V2.6-Flash cost $397,000 over 1 day 14 hours (over $10,000/hour), totaling about $31,000 per hour for both. Xiaomi stated the training expanded RL compute across three dimensions, supporting 1M context length and 2.7–3.7B tokens per step, with a multi-task training system covering Code, General, Visual, and Cyber directions. The company has open-sourced the technical report, training environment, and RL code.
Read source