Ant launches trillion-parameter thinking model Ring-2.6-1T: PinchBench scores 87.60, surpassing GPT

robot
Abstract generation in progress

CryptoWorld News: Ant Group’s Bailing large model team has launched the trillion-parameter thinking model Ring-2.6-1t (active parameters 6.3 billion), designed specifically for complex tasks and production environments. The model adds a new “dynamic thinking intensity” mechanism, flexibly balancing cognitive depth, token costs, and execution speed. The model offers two operating modes: high and xhigh. In agent mode (high), PinchBench scores 87.60, higher than GPT-5.4 xhigh and Gemini-3.1-pro high; Claweval scores 63.82. In deep thinking mode (xhigh), AIME 26 scores 95.83, and GPQA Diamond scores 88.27. The official says this mechanism is intended to reduce token spending and support high-frequency scenarios such as tool orchestration, programming, and multi-turn interactions. Starting today, the model will be jointly provided with Novita on the OpenRouter platform for a one-week free API trial (until May 15), and the weights will be open-sourced soon.

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
  • Pin