Ling-3.0-tiny

All projects

Apps & Tools

A fast 7.9B reasoning MoE model optimized for local inference on Macs and GPUs.

byAnt Ling

Playable

Ling-3.0-tiny media is blocked

Allow external media to connect to the provider and play this content.

Description

Ling-3.0-tiny is a lightweight hybrid reasoning Mixture-of-Experts (MoE) model with 7.9B total parameters and 1.3B activated parameters per token. It features a native 'thinking mode' for multi-step reasoning and is optimized for local deployment, achieving up to 90 tokens/s on Apple Silicon (M4 Pro) and 160 tokens/s on high-end GPUs.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.

Details

Why this is interesting: Impressive local performance (90 t/s on M4 Pro) for a reasoning MoE model with native thinking capabilities.
Attributed to: @AntLingAGI
Built with: Ling 3.0 Flash Fin Kling 3.0 Kling 3.0 Redwood AI Moonshine Tiny q8
Links: openrouter.ai
View tweet
Tags: #llm #reasoning #moe #local-ai

Discovered: X
First posted: X · Aug 7, 2026
Listed: Added 45d ago
Build: Strong
Models: Moderate

Prompt

/agent-browser use agent browser, access https://news.ycombinator.com/ . click to explore some different posts, write them to local markdown files, then generate a summary html with light and minimal style.

Source

Ling-3.0-tiny

Description

Timeline

Ling-3.0-tiny

Description

Timeline