Ant launches Ling-3.2-flash: A 1M context multimodal model free for two weeks

Ant Group just dropped a native multimodal model that eats text, images, and video in a 1M token window. It is free for two weeks.

Models · Source: TechNode

What happened

Ant Group officially released Ling-3.2-flash on November 5. This is a native multimodal artificial intelligence model designed to process multiple data types simultaneously. Users can input text, images, and video all at the exact same time. The model handles all of this within a massive one-million-token context window. You can drop entire textbooks, long lecture videos, and hundreds of photos into a single prompt.

The underlying architecture is massive but highly optimized for efficiency. The model contains 560 billion total parameters. However, it utilizes a sparse architecture that only activates 25 billion parameters per inference step. Ant Group states the model is specifically engineered for complex visual understanding and multistep agent tasks. It also targets highly specialized fields like 3D understanding, 3D interaction, and medical multimodal applications.

Developers do not need to learn a completely new ecosystem to start using this model. The system is available through a standard online chat interface for immediate testing. More importantly, the API is fully compatible with both OpenAI and Anthropic protocols. Ant Group is offering completely free access to the model for the first two weeks following the launch.

Key facts

Why it matters

Native multimodal capabilities completely change how engineers build data pipelines. In the past, you needed separate specialized models to transcribe a video, run optical character recognition on images, and parse the resulting text. Now, you can feed raw and mixed media directly into a single unified prompt. The one-million-token window allows you to process hours of video or hundreds of pages of PDF documents in one go. This drastically cuts down system latency and simplifies your entire backend architecture.

The API protocol compatibility is the most critical feature for fast-moving startups. Because Ling-3.2-flash supports both Anthropic and OpenAI standards, you can swap it into your existing tools immediately. The documentation explicitly notes integration support for developer tools like Claude Code, OpenClaw, Kilo Code, and Hermes Agent. Builders get a two-week window to test a heavy-duty 560-billion parameter model on their hardest multistep agent tasks. You get to benchmark top-tier multimodal performance without spending a single dollar on inference costs.

For builders

Swap into Claude Code for free

The API matches Anthropic protocols perfectly. You can plug this directly into Claude Code or your existing agent workflows today. You save significant money while testing a massive context window on your own codebase.

Consolidate your media pipelines

Stop chaining separate vision and text models together. Ling-3.2-flash handles text, photos, and video natively in one single API call. This reduces your infrastructure costs, lowers latency, and speeds up complex visual processing tasks.

Exploit the two-week free window

Ant Group is not charging for API access for the first fourteen days. Founders should run their most expensive, token-heavy medical or 3D data tasks right now. You get enterprise-grade inference without the massive cloud bill.

My take

I love when massive tech companies give away raw compute just to win market share. A one-million-token context window that eats video and plugs directly into Claude Code for free is an absolute gift for bootstrapped founders. Stop reading, swap your API keys, and run your heaviest data tasks before the free window closes.

Original reporting: TechNode. This is my rewrite and opinion.

More AI news for builders