📌 Repository Maintenance / Supplement

[yaotingwangofficial]

• This repository will be updated weekly.

• Welcome to supplement any missing information. Please leave your comments * in this issue * in the format of [paper title](github repo/arxiv url). We greatly appreciate your help!

Note:

• Given the rapid development of MCoT reasoning, we kindly ask any authors wishing to supplement their works into this survey to provide the relevant information in the [_specified format_].
• This will help us quickly iterate on the version to include the updated content and ensure its accuracy.
• We sincerely appreciate the researchers who have contributed to and provided feedback for this survey.

Best wishes to all!

---

Specified format:

(1) Benchmark:

Datasets	Year	Task	Domain	Modality	Format	Samples	With-or-Without-Rationale
ScienceQA	2022	VQA	Science	T, I	MC	21K	Yes

(2) Models:

Model	Foundational LLMs	Modality	Learning	Cold Start	Algorithm	Aha-moment
Deepseek-R1-Zero	Deepseek-V3	T	RL	❌	GRPO	✅
R1-Omni	HumanOmni-0.5B	T,I,V,A	SFT+RL	✅	GRPO	-

If possible, please provide the evaluation results on MMMU (Val), MathVista (mini), Math-Vision, and EMMA (mini).

[yaotingwangofficial]

[Done]

Survey:
Towards reasoning era: https://arxiv.org/abs/2503.09567

Reasoning model:
R3V: https://arxiv.org/pdf/2411.00855
R1-VL: https://arxiv.org/abs/2503.12937
VisRL: https://arxiv.org/abs/2503.07523

Benchmark:
OSWorld: https://arxiv.org/abs/2404.07972
AgentClinic: https://arxiv.org/abs/2405.07960