Ling-3.0-tiny: Lightweight Hybrid Reasoning MoE Model by inclusionAI

LinkinclusionAITool

inclusionAI has introduced Ling-3.0-tiny, a compact MoE model with 7.9B total parameters activating only 1.3B parameters per token. Utilizing a hybrid architecture of linear and latent attention mechanisms, it aims to deliver efficient reasoning and agentic workloads on local hardware.

Key points

  1. Features a 3:1 alternating stack of Kimi Delta Attention (KDA) and Multi-Head Latent Attention (MLA) for efficient long-context processing.
  2. Sparse MoE FFN with 128 experts, activating only 8 routed experts and 1 shared expert per token (1.3B active parameters).
  3. Native hybrid reasoning supporting fast outputs or multi-step thinking, configurable via enable_thinking.
  4. Optimised for local hardware including Apple Silicon (M4 Pro) and NVIDIA DGX Spark, using around 8.34 GiB VRAM at FP8 with an 8K context.
  5. Integration paths provided for inference runtimes like SGLang, vLLM, and an experimental Ollama build via MLX.

Why it matters

Advanced reasoning models often demand expensive data-centre GPUs. Ling-3.0-tiny highlights how hybrid architectural designs (KDA, MLA, sparse MoE) can bring multi-step reasoning and agentic workflows to local devices and edge hardware at manageable memory footprints.

Analysis

Ling-3.0-tiny reflects the continuing shift towards high-efficiency hybrid architectures. By pairing linear attention (KDA) with MLA and extreme MoE sparsity, inclusionAI keeps active parameters low without sacrificing reasoning depth. Demonstrating speeds up to 90 tokens/s on standard Apple Silicon machines, the model provides a viable path for deploying agentic routines and coding tasks locally without reliance on costly cloud infrastructure.

Mentioned links

Found in

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Article:Zhidx

    Ant Group open-sources compact MoE model Ling-3.0-tiny

    Ant Group's Bailian team has released Ling-3.0-tiny, an open-source lightweight mixture-of-experts model. It leverages DeepSeek's MLA and Kimi's KDA architectures with 7.9B total parameters, activating just 1.3B parameters per token for local edge deployment on hardware like Apple Silicon Macs and DGX systems.

    AI & LLMs· Announcement

  • Repository:inclusionAI/Ling

    Ling: Open-Source Mixture-of-Experts LLMs by InclusionAI

    InclusionAI has open-sourced Ling, a family of Mixture-of-Experts (MoE) large language models. The suite is offered in two sizes: Ling-lite with 16.8 billion total parameters (2.75 billion activated) and Ling-plus with 290 billion total parameters (28.8 billion activated).

    267starsPython

    AI & LLMs· Resource

  • Article:Thinking Machines Lab

    Thinking Machines Lab Releases Inkling-Small

    Thinking Machines Lab has introduced Inkling-Small, an open-weights Mixture-of-Experts model (276B total, 12B active parameters) designed to deliver performance comparable to Inkling at a fraction of its size and compute cost.

    AI & LLMs· Announcement

  • Link:drumih

    TurboFieldfare: Running Gemma 4 26B in 2 GB RAM on Apple Silicon

    TurboFieldfare is an open-source inference engine written in Swift and Metal that executes 4-bit Gemma 4 26B-A4B-IT on Apple M-series Macs using roughly 2 GB of RAM. While the model weights require 14 GB of disk space, the engine keeps only the shared model base and KV cache in memory and streams routed expert layers on demand directly from the SSD.

    AI & LLMs· Tool

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.