inclusionAI open-sources Ling-3.0-flash-VL: 124B MoE with 5.5B active and 1M context
jacek2023 · reddit · 2026-09-08
inclusionAI released Ling-3.0-flash-VL on Hugging Face:
- Sparse MoE with 124B total parameters, only 5.5B activated per token, supporting up to 1M-token context;
- Inherits Ling-3.0-flash's language, reasoning, and long-context strengths, adding native image and video understanding for real-world reasoning and agentic workflows;
- Architecture: ViT encoder + two-layer MLP projector; VideoRoPE encodes spatial position and temporal order for event localization, long-video QA, and clip editing; a 42-layer hybrid backbone alternates KDA and Gated MLA layers at a 5:1 ratio for efficient multimodal long-context processing.
More from Models
- Nat Lambert: Cohere's model is narrow RL training on internal infra ill-suited for open weights — natolambert · 2026-10-27
- OpenAI teases DOTS is coming to Europe 'soon' — mark_k · 2026-10-27
- Clifford Pickover challenges AI to decipher a message he encoded in Norse runes — pickover · 2026-10-27
- ChatGPT's crawler sample code strips out all Schema markup — VeryWellVersed · 2026-10-27
- Cohere sandbox: Qwen model tried to upload probe file to Beijing cloud server unprompted — simonguozirui · 2026-10-27
- Kimi K3.1 Pro tops open-weight models on FrontierML benchmark, still far behind Opus 5.5 — teortaxesTex · 2026-10-27