inclusionAI Releases Ming-3.0-audio-tiny Audio-Language Model
Start here
inclusionAI released Ming-3.0-audio-tiny, a native audio-language Mixture-of-Experts model with 9.3B total parameters, of which 2.7B are activated per inference step, under the MIT License. It takes audio plus text instructions and supports speech recognition, spoken question answering, audio understanding and reasoning, and optional synthesized speech through a Talker module, with Chinese and English as the primary languages. The release ships Transformers, vLLM-Omni and SGLang serving examples. The benchmark figures published alongside it also cover the larger Ming-3.0-audio-flash, which inclusionAI says reaches 99.8% on Artificial Analysis's Big Bench Audio; that score is reported for the flash model, not for the tiny one.
Generated from reports · updated 20 min ago
Report timeline
Follow the reports to see the event from different sides.
- inclusionAI ModelsinclusionAI releases Ming-3.0-audio-tiny, a 9.3B-parameter MoE audio-language model
inclusionAI released Ming-3.0-audio-tiny, a compact native audio-language Mixture-of-Experts model with 9.3B total parameters and 2.7B activated per inference step, under the MIT License. It accepts audio plus text instructions and supports speech recognition, spoken question answering, audio understanding and reasoning, and optional synthesized speech via a Talker module, with Chinese and English as primary languages. The release includes Transformers, vLLM-Omni and SGLang serving examples, and the shared benchmark figures also cover the larger Ming-3.0-audio-flash, which reaches 99.8% on Artificial Analysis's Big Bench Audio.
Heat for this event
Not enough continuous observations yet to draw a trend.