InternLM3-8B-Instruct
8B open model with a deep-thinking mode, trained on only 4T tokens, said to save over 75% of training cost versus similar-size LLMs.
Claims to beat Llama3.1-8B and Qwen2.5-7B on reasoning and knowledge tasks (OpenCompass). Apache-2.0.
- Date
- Wednesday, 15 January 2025
- Lab
- Shanghai AI Laboratory (InternLM)
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Pretraining tokens | 4T | company |
| Training-cost saving vs similar-size LLMs | over 75% company-claimed | company |
Date from the InternLM README news (2025.01.15). Cost-saving figure is the lab's own.
Sources
This record was checked against its sources on 6 October 2026. How we check