Intel / post

Ling-3.0-tiny opens weights in BF16, FP8 and INT4

Source: x.com
Date: Published Aug 11, 2026

Introducing the new member of the @AntLingAGI family, 👏Ling-3.0-tiny, delivering balanced coverage across reasoning, coding agents, instruction following and long-context tasks. 📣Now open weights on Hugging Face & ModelScope 🤗Hugging Face weights: 🔗BF16: https://t.co/QLuNy4YpDm 🔗FP8: https://t.co/Cq8KGCnihh 🔗INT4: https://t.co/2K4HNQQczO ⚙️ModelScope weights: 🔗BF16: https://t.co/DtlAWVk4zm 🔗FP8: https://t.co/rP1yPc2cVH 🔗INT4: https://t.co/XOdLocCxBo

Aug 11, 2026 View on X

Context

Ant Open Source released open weights for Ling-3.0-tiny, the smallest model in Ant's Ling family, aimed at reasoning, coding-agent, instruction-following and long- tasks. It shipped in three precision formats at once, BF16 (the full-precision format), FP8 and INT4 (progressively compressed formats that shrink memory use and speed up inference at some cost to accuracy), on both Hugging Face and ModelScope.

A companion Ant Ling post describes the same model in more technical detail: a 7.9-billion-parameter model with only 1.3 billion parameters active per token, using a 3:1 mix of two mechanisms and 128 sparse experts, with measured throughput on Nvidia's DGX Spark and on Apple silicon. Releasing all three precisions together, rather than only the full-precision version, means a team does not have to wait on a third party to produce a working build before deploying on constrained hardware.

Terms in this piece · Glossary

context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
quantization — Shrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.

Read the source x.com