New Models 2026-08-09

Ant Group's new model is 7.9 billion parameters and uses 1.3 billion of them at a time — free to try until Friday

InclusionAI, Ant Group's model unit, released Ling-3.0-tiny on August 6. It is a mixture-of-experts model with 7.9 billion total parameters that activates only about 1.3 billion for each token it produces, and it carries a 256,000-token context window with up to 32,000 tokens of output. Like its bigger sibling Ling-3.0-flash, released two weeks earlier at 124 billion parameters, it is a native hybrid-reasoning model: it can switch between an instant mode and a thinking mode rather than being one or the other. It supports native function calling and prompt caching, and Ant is aiming it at agents, instruction following and hardware-constrained deployment. It is already live on OpenRouter and on Vercel's AI Gateway, free of charge until August 14 at 08:00 Pacific time, where it took over the free slot Ling-3.0-flash had held. The open weights have not been published yet; Ant says they are coming.

Why it mattersThe interesting number here is not 7.9 billion, it is 1.3 billion. A model that only wakes a small fraction of itself per token costs far less to run and can fit on hardware that a frontier model cannot reach — a laptop, a phone, a machine on a factory floor with no reliable connection. Chinese labs have spent this year pushing that ratio further, and Ling-3.0-flash already claimed to match Ant's own trillion-parameter flagship at a fraction of the compute. Two cautions before you get excited: the weights are not out, so for now this is an API you rent rather than a model you own, and the free window closes on Friday. If you want to test whether a small local-class model is good enough for your workload, this week is the cheap time to find out.

#AI Agents

✓ Verified · 3 sources

Read in the app — free, in 9 languages

DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best
2026-08-21 Given four hours and a GPU to improve the way AI is trained, the best agent scored 0.25 out of 1 — and most never tried
2026-08-21 ChatGPT can now read your iMessages and send them — and the setting that lets it skip asking is the one OpenAI warns about
2026-08-21 Show the robot once — three to twelve seconds — and it gets the job right 59 times out of 100 with no training at all
2026-08-21 The agent invented a second person to vouch for its code. A 24-year-old in Texas refused to believe either of them.
2026-08-21

Ant Group's new model is 7.9 billion parameters and uses 1.3 billion of them at a time — free to try until Friday

Related stories