Ling-3.0-tiny – Ant Group's Ling native hybrid reasoning model Operational

Ling-3.0-tiny is a native hybrid reasoning model launched by Ant Group's Ling, with 7.9B total parameters and only 1.3B activated per token, optimized for mathematical reasoning, instruction following, and resource-sensitive deployment. The model natively supports tool calling and can handle mobile device and browser UI automation tasks, as well as fully local inference. It is currently available on OpenRouter and Vercel, and the model weights will be open-sourced soon.

Hybrid reasoning architecture: With 7.9B total parameters and only 1.3B activated per token, it delivers efficient computation for mathematical reasoning and instruction-following tasks.
Native tool calling: Built-in tool-use capabilities directly drive browser UI control and mobile device automation.
Fully local inference: Runs without cloud dependencies. When integrated with obsidian-cli, it can retrieve, organize, and generate content from local text repositories.
Lightweight text generation: Provides low-latency text generation for frequent scenarios such as local webpage translation and everyday task processing.

Sparse Mixture-of-Experts (MoE) architecture: The model has 7.9B total parameters, but a gating and routing mechanism activates only approximately 1.3B parameters per token. This preserves the expressive capabilities of a large model while compressing inference computation to the level of a small dense model.
Native hybrid reasoning paths: Multiple differentiated reasoning channels are built in. The model can dynamically select a lightweight, fast-response path or a deep chain-of-thought path based on task complexity, balancing mathematical reasoning accuracy with response speed for everyday instruction following.
Gated routing and load balancing: A learnable router assigns each token an optimal subset of experts, supplemented by a load-balancing loss function to prevent expert collapse and ensure that the 7.9B-parameter space is used efficiently and evenly.
Native tool-calling integration: Tool-use patterns are encoded directly into the model's representation space during pretraining or alignment, enabling the model to generate tool-call instructions that conform to tool interface specifications without external plugins, thereby automating browser UI and mobile device control.
On-device deployment optimization: Quantization and KV-Cache memory optimizations target local inference scenarios, enabling fully offline operation on consumer-grade hardware, eliminating cloud dependencies and reducing the network and computing costs of frequent API calls.

Visit a platform: Visit the OpenRouter official website (https://openrouter.ai) or the Vercel AI Gateway page.
Choose the model: Search for and open the details page for the inclusionai/ling-3.0-tiny:free model.
Obtain access: Register or sign in, then obtain an API key and call the service directly during the free trial period.
Wait for open sourcing: Follow Ant Ling's official channels and download the model files for local deployment once the weights are open-sourced.
Integrate tools: Configure obsidian-cli or browser automation plugins to enable local knowledge retrieval and UI control.

Extremely lightweight: It runs with only 1.3B activated parameters, with inference costs and GPU memory usage far below those of dense models of a similar scale, making it suitable for on-device and edge deployment.
Privacy first: It supports fully local operation, keeping data on the device and meeting the stringent data security requirements of scenarios such as finance and healthcare.
Native tool capabilities: External tools can be called without additional fine-tuning or plugins, lowering the development barrier for automation workflows.

Intelligent mobile assistants: Perform translation, question answering, and schedule management offline on mobile devices, responding without an internet connection.
Browser automation: Automatically execute webpage form filling, information extraction, and UI interaction tasks through native tool calling.
Local knowledge management: Integrate obsidian-cli to build a fully private knowledge base with local document retrieval and intelligent summarization.
Enterprise edge deployment: Deploy on industrial control computers or intranets to provide compliant AI services on resource-constrained devices.
High-frequency lightweight services: Run as a persistent backend interface to handle frequent requests such as webpage translation and content generation, significantly reducing computing costs.