Skip to main content
Rase AI Intelligence Newsletter

Models & assistants Model or AI service

Ling-1T publishes a trillion-parameter model with sparse activation

InclusionAI made Ling-1T instruction weights available to download. Its architecture has about one trillion total parameters and activates about 50 billion for each token. [1][2]

Model access
Downloadable original Ling-1T instruction model weights. [1][2]
License
MIT License for the published original Ling-1T checkpoint. [2]

What the model does

The team’s October 2025 report describes Ling-1T as the largest of three Ling 2.0 instruction models. The model card distinguishes it from the other checkpoints. [1][2]

Why it matters

A sparse model uses part of its capacity to process each token. That can affect the amount of computation required, while storing and serving the whole model is still a separate constraint. [1][2]

Who can use it

Infrastructure teams can evaluate the weights for text and instruction tasks with hardware suitable for the full checkpoint. Researchers can compare it with smaller Ling 2.0 sizes. [1][2]

Key dates

  1. Ling 2.0 report including Ling-1T first submitted [1]

Ways to use it

Compare a sample of ordinary text questions across versions and record memory use, response time and answer quality. Keep the serving settings constant for each comparison. [1][2]

Model limitations

The model card lists limitations with long-context tasks and multistep tool use. The number of active parameters is not a measure of the memory needed for the entire checkpoint. [2]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source InclusionAI’s first Ling 2.0 report covering Ling-1T
  2. Primary source inclusionAI/Ling-1T

Related profiles

Models & assistants

GPT-6.2 Astra and Sol

OpenAI

Astra targets demanding projects. Sol targets everyday tasks. Both support text and image input in the API.

High confidenceEvidence checked

Models & assistants

MAI Code 2 and Code 2 Flash

Microsoft AI

Code 2 targets complex software work. Flash targets frequent edits, explanations and routine fixes.

High confidenceEvidence checked

Models & assistants

MAI Realtime 1

Microsoft AI

Users can try a MAI Labs demo. Developers can register for a limited Foundry preview.

High confidenceEvidence checked