Back to news
launchAnt Group2026-11-06

Ant Ling released the native multimodal Ling-3.2-flash model on November 6 with 560B total parameters and API input pricing of 0.30 dollars per million tokens

Ling-3.2-flash has 560B total and 25B active parameters with a 1M context window, priced at 0.30 dollars per million input tokens and 0.90 dollars per million output tokens, free for the first two weeks.

On November 6, the Ant Ling team announced the release of the native multimodal model Ling-3.2-flash, with 560B total parameters, 25B active parameters and a 1M context window. The model natively fuses visual and language capability at the architecture level, accepting image, text and video input, and focuses on complex visual understanding, multimodal long-horizon agents, 3D understanding and interaction, and medical multimodality.

In the evaluation results Ant published, Ling-3.2-flash exceeds Anthropic Claude Opus 5 at max effort on three benchmarks: OmniDocBench1.5 scores 92.21 against 90.56, ZEROBench subset scores 43.71 against 39.52, and Vision2Web scores 85.86 against 80.20. In the DesignArena image-to-frontend arena the model enters the top ten with an Elo score of 1238, alongside competitors from Meta, DeepSeek and Anthropic.

On pricing, the Ling-3.2-flash API is 0.30 dollars per million input tokens and 0.90 dollars per million output tokens. The API is live through the Ant Ling platform, compatible with both the OpenAI and Anthropic protocols, and free for a limited trial during the first two weeks. The combination of 560B total and roughly 25B active parameters corresponds to a mixture-of-experts architecture where only part of the parameters are called in each forward pass.

Ling-3.2-flash蚂蚁百灵多模态定价开源