Neural Brew

Models & labs

Ant Ling Launches Ling-3.2-flash, a Native Multimodal 560B MoE with 1M Context

Ant Group has launched Ling-3.2-flash, a 560B/25B active MoE with a 1M token context, native multimodality (image, text, video), OpenAI- and Anthropic-compatible APIs, and free access for the first two weeks.

AI-generated portrait of LiorAI-generated portrait of Iris

By and

Published · 2 min read

A massive crystal prism focuses multiple colored light beams into a single luminous beam in the dark.
AI-generated illustration

Ant Ling, the model platform from Ant Group, launched Ling-3.2-flash on November 5, 2026, a native multimodal Mixture-of-Experts. The announcement was posted on November 5 on the @AntLingAGI account; the developer changelog confirms the same date. The model adopts the same footprint as Ling-3.1-flash-1019 — 560 billion total parameters, 25 billion active parameters, 1M context — but adds integrated image + text + video input directly into the architecture, without an external vision module.

Four Focus Areas, Open API, Free for Two Weeks

The models page highlights four areas: complex visual reasoning, long-horizon multimodal agents, 3D understanding and interaction, and medical multimodality (imaging and report interpretation, presented as professional analysis aids — not as standalone clinical tools).

In terms of access, Ling-3.2-flash is available via web chat and an API compatible with OpenAI and Anthropic. The changelog and the pricing page indicate free access for the first two weeks following launch, after which it will revert to the “original price.” That original price is not listed for Ling-3.2-flash on the page reviewed on November 6: we cannot invent a rate.

Following Ling-3.1, No Open Weights (Yet)

Ling-3.1-flash-1019 was released under the MIT license; we covered it in October. For Ling-3.2-flash, as of November 6, neither the changelog nor the model documentation mentions an open-weight release: the model is served via API and chat. Ling-3.0-flash-VL, a smaller model (124B / 5.5B active), remains the open multimodal reference cited in the same documentation.

Ant Ling also claims, on X, scores exceeding “Opus 5 (max)” on OmniDocBench 1.5, ZEROBench(Sub), and Vision2Web. These are in-house lab numbers, without independent verification at the time of our publication.

Our take

Ling-3.2-flash complements rather than replaces Ling-3.1-flash: identical MoE size, identical one-million-token window, but vision and video are now native. The medical angle is announced without clinical specifics — it should be treated as a stated capability, not a healthcare device. The contrast with the previous generation is stark regarding openness: Ant released Ling-3.1-flash-1019 under MIT; 3.2 arrives first via a temporary free API. For developers, the two-week window offers a chance to test native multimodality; judging the price-to-performance ratio will require waiting for the actual rate card and, potentially, an open-weight release that has not been announced at this stage.

Sources

  1. Ant Ling — Changelog (November 5, 2026) Primary source
  2. Ant Ling — Ling Models Documentation Primary source
  3. Ant Ling — Product Website Primary source
  4. Ant Ling (X) — Ling-3.2-flash Announcement Primary source
  5. Ant Ling — Pricing Primary source — Confirms 2 weeks of free access; no post-promotional rate listed for Ling-3.2-flash
  6. Hugging Face — inclusionAI/Ling-3.1-flash-1019 Secondary source — Context: October 27 MIT open release for the previous generation

A primary source comes directly from the company, lab or organization concerned. A secondary source reports the information without being its author: we flag it as such. Spotted an error? See our corrections policy.