Ling-1T: A Trillion-Parameter Approach to Efficient AI Reasoning

Author(s): Gowtham Boyina

Originally published on Towards AI.

Exploring InclusionAI’s Claims of Balancing Scale with Computational Efficiency

The AI landscape continues to evolve rapidly in 2025, with new models pushing various boundaries. Ling-1T, released by inclusionAI, represents an ambitious entry into the trillion-parameter space. The team claims their flagship non-thinking model achieves competitive performance while maintaining efficiency — though independent verification of these claims through third-party benchmarks remains pending.

Ling-1T introduces a novel architecture with 1 trillion total parameters, of which about 50 billion are activated per token, achieving computational efficiency through a Mixture-of-Experts (MoE) architecture. The training methodology emphasizes sparse activation, mixed-precision training, and a unique reasoning process called Evolutionary Chain-of-Thought (Evo-CoT). Reported benchmark results claim high accuracy on various tasks, aiming to position Ling-1T competitively among both open-source and closed-source models while highlighting a commitment to transparency and independent validation of its performance metrics.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI