Views
No views yet
Qwen3_5MoeForConditionalGeneration). Intermediate stage before mid-training and RL.| Benchmark | Metric | Score |
|---|---|---|
| BrowseComp | avg@3 | 45.1 |
| Mind2Web 2 | avg@3 | 26.5 |
| HLE | avg@3 | 39.49 |
| DeepResearch Bench | avg@3 | 36.35 |
| BrowseComp-Plus | avg@3 | 57.9 |
| WideSearch | Item F1 avg@4 | 61.1 |
| GAIA | avg@3 | 83.5 |
| LiveResearchBench | avg@3 | 64.69 |
| Type | Resources |
|---|---|
| 35B checkpoints | RL, MT+SFT, MT, SFT |
| 30B checkpoints | RL, MT+SFT, SFT |
| Smaller checkpoints | 9B, 4B, 2B |
| Training data | RL data, SFT objective data, SFT open-ended data, Mid-training data |
1from transformers import AutoModelForCausalLM, AutoTokenizer
2
3model_id = "osunlp/QUEST-35B-SFT"
4tokenizer = AutoTokenizer.from_pretrained(model_id)
5model = AutoModelForCausalLM.from_pretrained(
6 model_id, device_map="auto", torch_dtype="auto",
7)tokenizer.apply_chat_template(...) before passing prompts.1@misc{xie2026quest,
2 title={QUEST: Training Frontier Deep Research Agents with Fully Synthetic Tasks},
3 author={Xie, Jian and Lin, Tianhe and Wang, Zilu and Ning, Yuting and Yao, Yuekun and Xue, Tianci and Zhang, Zhehao and Li, Zhongyang and Zhang, Kai and Wu, Yufan and Chen, Shijie and Gou, Boyu and Han, Mingzhe and Wang, Yifei and Lee, Vint and Wei, Xinpeng and Wang, Xiangjun and Su, Yu and Sun, Huan},
4 journal={arXiv preprint arXiv:2605.24218},
5 year={2026}
6}