inclusionAI
5 storiesinclusionAI is an AI group that releases open-weight models, mostly under the Ling and Ming names, on Hugging Face. Recent coverage has centered on releases: Ling-3.1-flash-1019 weights and a BF16 base model, keeping 560B total and 25B active parameters with a 1M context and raising its AA Intelligence Index score from 41 to 44; Ming-3.0-audio-flash, a 126B audio-language model with 6.5B activated parameters; the 9B Realtime-Venus omni models; and Ling-3.0-flash-VL at 124B parameters.
Related topics
Ling-3.1-flash-1019 weights are now on Hugging Face, five days after a "coming soon" teaser. The model keeps 560B total and 25B active parameters with a 1M context, and scores 44 on AA Intelligence Index v4.3.2, up from 41, with AutomationBench-AA rising from 62% to 68%. A BF16 base model is also released for further training, though the 1019 README's SGLang example still points to the older model path.
inclusionAI released Ming-3.0-audio-flash, a native audio-language Mixture-of-Experts model with 126B total parameters and 6.5B activated per inference. It handles Chinese and English speech recognition, audio understanding and reasoning, with optional synthesized speech, and is available under the MIT license on Hugging Face and ModelScope.
inclusionAI's Ling-3.1-flash appeared on Hugging Face with 560B total parameters, 25B active per token and a 1M-token context window, a large jump from the 127B Ling-3.0.
inclusionAI released two Realtime-Venus checkpoints. The 9B Realtime-Venus-Omni, adapted from MiniCPM-o 4.5, continuously watches and listens, decides when to respond, and generates text and speech on a shared causal timeline. Realtime-Venus-Audio focuses on audio understanding and conversation. Both support proactive interaction, semantic interruption, and training-free long-video memory, with an async harness on GitHub.
inclusionAI published Ling-3.0-flash-VL on Hugging Face, a multimodal model with native image and video understanding, a 1M-token context window, and 124B total parameters. It activates about 5.5B parameters per token and uses sparse MoE and a hybrid backbone for efficient long-context multimodal reasoning.
That is everything
The A.I. Daily
Ling-3.1-flash-1019 weights and base model released