Ant Group open-sources Ling-3.0-flash hybrid reasoning model

Ant Group's Bailing large model team announced that its new native hybrid reasoning model Ling-3.0-flash is now officially open-sourced. The model uses a MoE architecture with 124B total parameters and 5.1B activated parameters, and is available in FP8, FP4, INT4 versions, supporting API calls, on-premises deployment, and high-performance deployment. On specified GPU configurations, it can reach an average output speed of over 1100 tokens per second.

On the Artificial Analysis leaderboard, Ling-3.0-flash achieves an output speed of 353 tokens per second, with an average weighted cost of about $0.04 per task and an average weighted decoding time of about 1.4 minutes, showing advantages in both intelligence-cost and intelligence-time. The company said that from August 7 to August 31, 2026, the model will be available at a 75% discount on Ling Studio, with prices returning to normal from September 1. The open-source links are available on Hugging Face and ModelScope.