INTELLECT-3.1
INTELLECT-3.1 continues training INTELLECT-3 on mathematics, coding, software engineering, and agent tasks. It retains the 106-billion-parameter mixture-of-experts design with 12 billion active parameters.
What the release provides
The model card provides a vLLM serving example on two H200 GPUs. It selects the qwen3_coder tool-call parser and deepseek_r1 reasoning parser. This is a documented setup, not a universal minimum hardware requirement.INTELLECT-3.1 official model card
How to use it well
Practical guidance. Keep reasoning and tool-call parsing separate in your application. Confirm that generated tool arguments match the schema before executing them.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Context window | 131,072 tokens Configured position limit. Serving limits depend on the runtime and settings. INTELLECT-3.1 model configuration |
| Base ancestry | GLM-4.5-Air-Base, through INTELLECT-3INTELLECT-3.1 official model card |
| Repository license | MITINTELLECT-3.1 official model card |
Related models
- INTELLECT-3
Prime Intellect ยท INTELLECT
A 106-billion-parameter mixture-of-experts reasoning model with 12 billion active parameters.
Launch date not confirmedDownloadable