Activating only 5.1 billion parameters, Ant’s new model’s 11-test suite surpasses the trillion-level flagship

ME News message, July 24 (UTC+8). According to Beating monitoring, Inclusion AI (Ant Bailing) has released Ling-3.0-flash. The model has 124 billion parameters in total, but only 5.1 billion activated parameters. Based on the comparison officially released, in 12 benchmarks it exceeds Ling-2.6-1T in 11 of them, with activated parameters of only about one twelfth of this trillion-level flagship. Architecturally, it combines Kimi Delta Attention and MLA attention layers in a 5:1 ratio to strengthen long-text memory. The model natively supports a 256K context and can scale up to 1 million. Ling-3.0-flash is already live on the Ant Bailing API and OpenRouter, and supports two modes: thinking and non-thinking. Currently, the model weights have not been opened, and the performance comparisons mainly come from Ant Bailing’s official tests; third-party evaluations are still needed for verification. (Source: BlockBeats)
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned