{"code":0,"message":"成功","data":{"id":1,"sectionKey":"about","title":"About Page Configuration","titleEn":null,"content":{"founders":[{"alt":"THUNLP","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/founder-thunlp.png"},{"alt":"ModelBest","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/founder-modelbest.svg"}],"milestones":[{"date":"2026.06","text":"In June, PilotDeck desktop version was officially launched, supporting multi-terminal use on Windows and macOS, and completed multiple experience upgrades around stability, setting interface, IM channel interaction and model access."},{"date":"2026.05","text":"In May, the community open sourced MiniCPM-V 4.6 and MiniCPM5-1B; PilotDeck was released in the same period, providing on-device model integration, multi-agent task management, and continuous operation."},{"date":"2026.04","text":"In April, the community open sourced VoxCPM 2. The 2B speech foundation model supports multiple languages, dialects, timbre design, and timbre control, with native 48kHz audio output."},{"date":"2026.03","text":"In March, the community open sourced the EdgeClaw device-cloud collaborative agent framework, which supports privacy and cost configuration through a three-tier security mechanism and a cost-aware collaboration mechanism."},{"date":"2026.02","text":"In February, the community open sourced MiniCPM-o 4.5, which uses a native full-duplex omni-modal architecture and supports streaming interaction across video, audio, and speech output; UltraData tiered data management was released in the same period."},{"date":"2026.01","text":"In January, the community open sourced ChatDev 2.0, AgentCPM-Explore, AgentCPM-Report and UltraRAG 3.0, covering zero-code multi-agent construction, in-depth exploration of client-side agents, localized in-depth research writing and white-box RAG development, further enriching the ecosystem of client-side agents and scientific research tool chains."},{"date":"2025.11","text":"In November, UltraRAG 2.1 was updated and released, which strengthens knowledge access, native multi-modal support and unified evaluation system. It also adds VisRAG Pipeline to realize a complete closed loop of local PDF database construction, multi-modal retrieval and generation."},{"date":"2025.09","text":"In September, the community released MiniCPM 4.1 and VoxCPM, which introduced efficient deep thinking capabilities into the device-side text model and open-sourced a high-fidelity speech generation base model to expand the capabilities of the device-side model in reasoning and voice interaction."},{"date":"2025.08","text":"In August, the community open sourced MiniCPM-V 4.0 and MiniCPM-V 4.5 and published their OpenCompass results: an average of 69.0 for MiniCPM-V 4.0 and 77.0 for MiniCPM-V 4.5; UltraRAG 2.0 was released in the same period."},{"date":"2025.06","text":"In June, the community open sourced MiniCPM 4.0, a natively sparse on-device model, together with its sparsity and inference-speed experiment results."},{"date":"2025.01","text":"In January, the community open sourced MiniCPM-o 2.6, supporting streaming of speech, text, images, and video, plus background-sound recognition, timbre conversion, voice cloning, and streaming video understanding. It scored an average of 70.2 on OpenCompass across 8 mainstream benchmarks."},{"date":"2024.11","text":"In the second half of the year, the MiniCPM device-side model ecosystem accelerated its expansion and was integrated and adopted by industry partners such as Huawei Cloud, MediaTek, Intel, Baidu Smart Cloud, Great Wall Motors, and Accelerated Evolution Robots. Its implementation scenarios cover smart cockpits, robots, AI Phones, AI PCs, and device-cloud collaboration."},{"date":"2024.09","text":"In September, the community open sourced MiniCPM 3.0. On a composite evaluation of 15 benchmarks including MMLU, HumanEval+, and BFCL v2, it averaged 66.3, against 61.0 for the GPT-3.5-Turbo-0125 reference."},{"date":"2024.08","text":"In August, the community open sourced the MiniCPM-V 2.6 end-side multi-modal model, which fully benchmarked against GPT-4V on the end-side and set a new end-side multi-modal performance record."},{"date":"2024.07","text":"In July, the community released the IoA (Internet of Agents) research framework for studying collaboration among multiple agents."},{"date":"2024.02","text":"The community open sourced the first version of the MiniCPM series, built for on-device deployment; model parameters, runtime requirements, and evaluation results are given in the technical report."},{"date":"2023.06","text":"The community released three open-source projects — XAgent, AgentVerse, and ChatDev — covering single-agent intelligence, multi-agent intelligence, and agent application frameworks respectively. As of August 2026, their GitHub star counts were 8,544, 5,115, and 34,126."},{"date":"2022.07","text":"In July 2022, the community launched the free course \"OpenBMB Large Model Open Course\"."},{"date":"2022.04","text":"The OpenBMB open source community was officially established. It was jointly initiated by Tsinghua University's THUNLP Laboratory and Modelbest to establish the community mission of \"let large models fly into thousands of households\"."},{"date":"2022 -\n2023","text":"The community has successively open sourced CPM-1, CPM-2, CPM-3, CPM-Ant, CPM-Bee and other series of models, and continues to promote the open source ecological construction of Chinese large models."},{"date":"2020.05","text":"The core team of the community took part in the \"WuDao\" large model project and released CPM-1, a 2-billion-parameter Chinese model, in 2020."},{"date":"2018.08","text":"Tsinghua University's THUNLP Lab, an early core team of the community, released the knowledge-guided pre-training model ERNIE in 2018, laying the research foundation for the CPM series."}],"memberLogos":[{"alt":"THUHCSI","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/member-thuhcsi.svg"},{"alt":"Open knowledge graph","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/member-2.png"},{"alt":"BIAA","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/member-biaa.svg"},{"alt":"Member units","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/member-gold.svg"},{"alt":"Member units","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/member-5.svg"},{"alt":"Member units","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/member-university.svg"},{"alt":"Northeastern University","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/member-db.svg"},{"alt":"Zhihu","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/member-8.png"}],"chipPartners":[{"alt":"Intel","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/chip-intel.svg"},{"alt":"MediaTek","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/chip-mediatek.png"},{"alt":"NVIDIA","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/chip-nvidia.svg"},{"alt":"Qualcomm","src":"https://synotag.oss-cn-beijing.aliyuncs.com/openbmb/public/images/about/chip-qualcomm.png"}]},"contentEn":null,"sortOrder":10,"published":true}}