inclusionAI/Ling-1T
Hugging Face
TEXT GENERATIONConcurrency Cost:4Model Size:1000BQuant:FP8Ctx Length:32kPublished:Oct 2, 2025License:mitArchitecture:Transformer0.5K Open Weights Warm

Ling-1T, developed by inclusionAI, is a flagship non-thinking model in the Ling 2.0 series, featuring 1 trillion total parameters with approximately 50 billion active parameters per token. Pre-trained on over 20 trillion high-quality, reasoning-dense tokens, it supports up to 128K context length and utilizes an evolutionary chain-of-thought (Evo-CoT) process. This model excels in efficient reasoning, complex problem-solving across code, math, and logic, and demonstrates strong capabilities in visual reasoning and front-end code generation, rivaling closed-source APIs in performance while prioritizing efficiency.

Loading preview...

Ling-1T: A Trillion-Parameter Model for Efficient Reasoning

Ling-1T, developed by inclusionAI, is the first flagship non-thinking model in the Ling 2.0 series, boasting 1 trillion total parameters with approximately 50 billion active parameters per token. It is pre-trained on over 20 trillion high-quality, reasoning-dense tokens and supports up to 128K context length. A key differentiator is its Evolutionary Chain-of-Thought (Evo-CoT) process, which significantly enhances reasoning efficiency and depth, allowing it to balance accuracy and performance on complex benchmarks.

Key Capabilities

  • Flagship-Level Efficient Reasoning: Consistently demonstrates superior complex reasoning across code generation, software development, competition-level mathematics, professional math, and logical reasoning, outperforming many open-source and closed-source models.
  • Aesthetic Understanding & Front-End Generation: Excels in visual reasoning and front-end code generation, combining deep semantic understanding with precise code synthesis, and ranks first among open-source models on ArtifactsBench.
  • Emergent Intelligence at Trillion-Scale: Exhibits strong emergent reasoning and transfer capabilities, achieving ~70% tool-call accuracy on BFCL V3 with minimal instruction tuning.
  • Advanced Training: Built on the Ling 2.0 architecture for trillion-scale efficiency, utilizing innovations like 1/32 MoE activation ratio, MTP layers, and is the largest FP8-trained foundation model to date.

Good For

  • Applications requiring highly efficient and accurate complex reasoning.
  • Code generation and software development, particularly for front-end code with aesthetic considerations.
  • Mathematical problem-solving at competitive and professional levels.
  • Developing general, collaborative human-AI intelligence systems.
  • Tasks benefiting from long context understanding and advanced chain-of-thought processing.
Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p