AnswersPublished25 sources
Ant Group’s Ling-3.2-Flash: A 560B Multimodal Model with 1M Context
Ling 3.2 flash is Ant Group’s native multimodal model with 560 billion total parameters, 25 billion active parameters and a 1 million token context window. Ant positions it for complex visual understanding, long horizon multimodal agents, 3D interaction and medical multimodal tasks.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Ling 3.2 flash is Ant Group’s native multimodal model with 560 billion total parameters, 25 billion active parameters and a 1 million token context window.
- Ant positions it for complex visual understanding, long horizon multimodal agents, 3D interaction and medical multimodal tasks.
- Ant’s documentation confirms a two week free launch offer and chat access, with APIs compatible with OpenAI and Anthropic protocols.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Ant Group announce about Ling-3.2-flash on November 6, 2026, including its native architectural integration of image, text and vide. Article summary: Ant Group announced Ling-3.2-flash as a native multimodal flagship with 560 billion total parameters, 25 billion active parameters and a 1-million-token context window, emphasizing visual reasoning and multimodal agents.. Topic tags: general, documentation, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks,
Ant Group’s Ling-3.2-flash is a large multimodal model designed to take image, text and video input, with a 1-million-token context window. Ant lists 560 billion total parameters and 25 billion active parameters, and highlights visual understanding and multimodal agents as central use cases. The benchmark results released around launch are promising, but they are Ant-reported comparisons on a small set of tests—not proof of broad superiority. 1
3
6
20
Launch coverage appeared on November 6, while Ant’s model and API documentation carry November 5 dates. The available sources therefore differ by a day on the announcement timing. 1
3
6
What Ling-3.2-flash offers
Ant describes Ling-3.2-flash as natively multimodal: image, text and video are supported as inputs, including through its OpenAI-compatible API. The model is listed with a context window of up to 1 million tokens, 560 billion total parameters and 25 billion activated parameters. 1
3
Launch coverage describes the model as a mixture-of-experts (MoE) system. In that design, the active-parameter figure refers to the portion used at a time; it is not the model’s total size. The listed figures are therefore 560B total and 25B active—not a 25B-parameter model. 7
10
Ant names four focus areas: complex visual understanding, multimodal agents that handle long-horizon tasks, 3D understanding and interaction, and medical multimodal work. These are the company’s stated priorities, rather than independent evaluations of performance in each field. 1
6
How it compares with earlier Ling models
Ant’s documentation lists Ling-3.1-flash and Ling-3.0-flash-VL alongside Ling-3.2-flash. It also describes Ling-3.0-flash-VL as a native multimodal model supporting image, text and video input. So Ling-3.2-flash should not be described as the first Ling model to support video. 2
5
Ant’s materials position Ling-3.2-flash as the latest-generation model and give its 1M context and 560B/25B parameter figures. The supplied evidence does not provide a complete, like-for-like specification comparison with Ling-3.1-flash, so claims about precisely how much it improves on that model should be treated cautiously. 1
2
Ant’s benchmark claims—and what they show
In results attributed to Ant, Ling-3.2-flash scored higher than Claude Opus 5 (max) on three benchmarks. These are vendor-reported results, not independently verified comparisons in the materials available here. 6
20
| Benchmark | Ling-3.2-flash | Claude Opus 5 (max) |
|---|---|---|
| OmniDocBench 1.5 | 92.21 | 90.56 |
| ZEROBench subset | 43.71 | 39.52 |
| Vision2Web | 85.86 | 80.20 |
The results support a narrower takeaway: Ling-3.2-flash performed well on these reported tests, which include document and visual-to-web tasks. They do not establish how it performs across other models’ evaluations or across general reasoning tasks. 6
20
Separately, launch coverage reported that Ling-3.2-flash placed in DesignArena’s top 10 for image-to-frontend generation with an Elo score of 1,238. The leaderboard also lists the model at 1,238. 6
25
Access, free offer and pricing caveat
Ant’s documentation says Ling-3.2-flash is available through a chat interface and that its API supports OpenAI- and Anthropic-compatible protocols. Ant also confirms a free-access offer for the first two weeks after launch. 4
27
28
Some launch coverage quotes prices of $0.30 per million input tokens and $0.90 per million output tokens. However, the supplied excerpt from Ant’s official pricing page confirms the free period but does not establish those regular rates for Ling-3.2-flash. Treat the quoted figures as reported prices, not as verified by that official pricing excerpt. 7
15
27
Are Ling-3.2-flash’s weights open?
The launch coverage provided here describes Ling-3.2-flash’s weights as unreleased at launch. It also reports that Ling-3.1-flash weights were available under an MIT license, but that does not establish that Ling-3.2-flash will receive the same license—or any open-weight release. 11
16
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Ant Group’s Ling-3.2-Flash: A 560B Multimodal Model with 1M Context"?
Ling 3.2 flash is Ant Group’s native multimodal model with 560 billion total parameters, 25 billion active parameters and a 1 million token context window.
What are the key points to validate first?
Ling 3.2 flash is Ant Group’s native multimodal model with 560 billion total parameters, 25 billion active parameters and a 1 million token context window. Ant positions it for complex visual understanding, long horizon multimodal agents, 3D interaction and medical multimodal tasks.
What should I do next in practice?
Ant’s documentation confirms a two week free launch offer and chat access, with APIs compatible with OpenAI and Anthropic protocols.
Sources
developer.ant-ling.comLing
news.aibase.comAnt Bailing Releases Ling 3.2 Flash: Surpasses Claude Opus 5 on Three Core Benchmarks, Free for the First Two Weeks
ithome.com蚂蚁百灵大模型Ling 3.2 Flash 发布：三项基准超越Claude ...