ModelBest
A Chinese AI startup focused on the research and development of efficient small-scale models, known for its MiniCPM series of edge-side language models.
Timeline (last 90 days)
The ModelBest team created the first 2B-parameter independent discrimination model, enabling on-device parsing to automatically correct errors without manual intervention.
ModelBest officially open-sourced the on-device multimodal model MiniCPM-OC2, which took multiple global first places on document benchmarks with an extremely lightweight parameter count of only 1.3B.
Tsinghua University's Sang Doupi Lab, together with teams including ModelBest and OpenManB, open-sourced the agent operating system Pilotech 2.0.
ModelBest simultaneously released the Self-Replay Fine-Tuning method, aiming to expand model capabilities at low cost while reducing forgetting of原有 capabilities.
ModelBest (OpenBMB) released MiniCPM-V 4.7, the newest member of the MiniCPM-V series, including 35B-A3B, the series' first MoE architecture variant.
ModelBest's edge model MiniCPM 5 2B has been adapted for OpenVINO and can perform local Agent tasks such as tool calling on-device.
The MiniCPM series is an open-source small model launched by ModelBest, focused on edge-side deployment and high cost-effectiveness.
ModelBest's MiniCPM5 2B ranked second on the trending chart with 67.6K daily downloads and 1.17K likes, targeting on-device and edge deployment and supporting a 128K context.
All Facts (7)
The ModelBest team created the first 2B-parameter independent discrimination model, enabling on-device parsing to automatically correct errors without manual intervention.
50%UnverifiedModelBest's edge model MiniCPM 5 2B has been adapted for OpenVINO and can perform local Agent tasks such as tool calling on-device.
50%UnverifiedThe MiniCPM series is an open-source small model launched by ModelBest, focused on edge-side deployment and high cost-effectiveness.
50%UnverifiedModelBest officially open-sourced the on-device multimodal model MiniCPM-OC2, which took multiple global first places on document benchmarks with an extremely lightweight parameter count of only 1.3B.
50%UnverifiedTsinghua University's Sang Doupi Lab, together with teams including ModelBest and OpenManB, open-sourced the agent operating system Pilotech 2.0.
50%UnverifiedModelBest simultaneously released the Self-Replay Fine-Tuning method, aiming to expand model capabilities at low cost while reducing forgetting of原有 capabilities.
50%UnverifiedModelBest (OpenBMB) released MiniCPM-V 4.7, the newest member of the MiniCPM-V series, including 35B-A3B, the series' first MoE architecture variant.
50%