Ant Group's inclusionAI Open-Sources Ling-3.1-flash: 560B Params, 1M Context, MIT License

Ant Group's inclusionAI open-sourced Ling-3.1-flash on Hugging Face on October 14. It has about 560B total parameters, roughly 25B activated per token, and a context window of up to 1M tokens — a big jump over the previous generation; jacek2023 noted that version 3.0 had only 127B total parameters. The model targets real-world development work, everyday office tasks, long-horizon software engineering, and complex Agent tasks, with Day-0 support from vLLM.

Confirmed

Weights are on Hugging Face (inclusionAI/Ling-3.1-flash): about 560B total parameters, 25B activated per token, up to 1M token context.
According to Zhidongxi, the model was released by Ant's Bailing team on September 30 and officially open-sourced on October 14; third-party Artificial Analysis has already run hands-on tests on office and Agent tasks.
Official benchmark charts show its performance on benchmarks like GDPval-AA and Skill (as relayed by CiobanuXashi).
Ant's Ling team showed a demo turning rough requirements directly into UI and interactive prototypes.
The vLLM project announced Day-0 support for Ling-3.1-flash.

Why it matters

Going from 127B to 560B parameters in a single model, paired with 1M context, shows Chinese tech giants racing toward extra-large parameters and long context in open source, natively aimed at Agent and real-world development scenarios rather than leaderboard chasing.
Open weights plus vLLM Day-0 support sharply lower the deployment barrier for the community; BerryBlushRogers691 noted users are already planning to run it locally on 4 DGX Spark machines, reflecting interest in its local playability.

2026-10-14 ~ 2026-10-15 · 8 related posts

Primary sources

Ant Group Open-Sources Ling-3.1-flash: 560B Hybrid MoE with 1M Context, MIT License — AdinaYakup · 2026-10-15 15:25 (UTC+08:00)
Ant's Ling-3.1-flash goes open source: 560B params, 25B active, 1M context, vLLM day-0 — vllm_project · 2026-10-14 23:39 (UTC+08:00)
Ant's Ling-3.1-flash goes open source: 560B params, 25B active, 1M context — 智东西 · 2026-10-15 13:19 (UTC+08:00)

inclusionAI's Ling-3.1-flash jumps to 560B params with 1M context, up from 127B — jacek2023 · 2026-10-14
[source] Ant's Ling-3.1-flash goes open source: 560B params, 25B active, 1M context, vLLM day-0 — vllm_project · 2026-10-14
Ling-3.1-flash Goes Open Source: 560B Params, 25B Active, 1M Context — CiobanuXashi · 2026-10-14
Can 4 DGX Sparks Run Ling-3.1-flash? Redditor Weighs 560B-A25B MoE at Home — BerryBlush_Rogers691 · 2026-10-14
[source] Ant's Ling-3.1-flash goes open source: 560B params, 25B active, 1M context — 智东西 · 2026-10-15
Ant's Ling releases open-source Ling-3.1-flash: 560B-param MoE built for real-world agent workflows — 量子位 · 2026-10-15
[source] Ant Group Open-Sources Ling-3.1-flash: 560B Hybrid MoE with 1M Context, MIT License — AdinaYakup · 2026-10-15
Ling-3.1-flash hits Hugging Face with thinking toggle and tool-call template — AdinaYakup · 2026-10-15