Qwen3 1.7B Base
Alibaba · 2025-04-29 · 1.7B parameters
Qwen3-1.7B-Base is the pre-trained (not instruction-tuned) 1.7B dense checkpoint of Alibaba's Qwen3 family, released April 29, 2025 under Apache 2.0 alongside the post-trained Qwen3-1.7B. Qwen describes a three-stage pre-training on about 36 trillion tokens in 119 languages, ending with long-context training at 32,768 tokens. It has 1,720,574,976 distinct parameters, counting the tied input and output embedding matrix once.
Benchmark scores
| Benchmark | Score |
|---|---|
| BBH (BIG-Bench Hard) | 54.5 |
| Capability | 101.0 |
| GSM8K | 75.4 |
| MATH | 43.5 |
| MGSM | 50.7 |
| MMLU | 62.6 |
| MMLU Pro | 36.8 |
| MMLU-Redux | 61.7 |
| SuperGPQA | 20.9 |