Original Title: Ant Group Bailing Large Model Ling 3.2 Flash Released: Outperforms Claude Opus 5 on Three Benchmarks, Limited-Time Free Access for First Two Weeks
According to IT Home, the Ant Group Bailing team has launched the native multimodal model Ling-3.2-flash, featuring 560B total parameters, approximately 25B active parameters, and a 1M context window. It supports image, text, and video inputs, with a focus on enhancing complex visual understanding, long-horizon multimodal agents, 3D understanding and interaction, and medical multimodality. The official announcement states that it outperforms Claude Opus 5 (max) on three benchmarks: OmniDocBench1.5, ZEROBench (Sub), and Vision2Web. The API is now live, with pricing at 0.30 USD per million input tokens and 0.90 USD for output. It is compatible with OpenAI and Anthropic protocols, and offers a limited-time free period for the first 2 weeks after launch.
Why it matters The official announcement claims superior performance on three visual benchmarks compared to Claude Opus 5, and compatibility with both OpenAI and Anthropic protocols ensures low switching costs.
Credibility: Fairly credible · Source: IT Home · Event time: 11-06 08:00 (Beijing Time)
Reasoning
- IT Home reports the official release, including official evaluations and pricing details
Risks
- The benchmark results are published by Ant Group officially and have not yet been reproduced by third parties