---
license: apache-2.0
language:
- en
tags:
- spatial-reasoning
- multimodal
- vision-language
- scene-graph
- reinforcement-learning
- mixture-of-experts
base_model: Qwen/Qwen3-VL-30B-A3B-Instruct
pipeline_tag: image-text-to-text
---

# SpatialThinker-30B

<p align="center">
  <a href="https://arxiv.org/abs/2511.07403">
    <img src="https://img.shields.io/badge/arXiv-2511.07403-b31b1b.svg" alt="arXiv">
  </a>
  <a href="https://hunarbatra.com/SpatialThinker">
    <img src="https://img.shields.io/badge/🌐%20Project%20Page-blue.svg" alt="Project Page">
  </a>
  <a href="https://github.com/hunarbatra/SpatialThinker">
    <img src="https://img.shields.io/badge/GitHub-Repository-black.svg" alt="GitHub">
  </a>
</p>

**SpatialThinker-30B** is a 30B-parameter Mixture-of-Experts (3B active) multimodal large language model trained with reinforcement learning to integrate structured spatial grounding with multi-step reasoning. It scales the SpatialThinker method to the Qwen3-VL-30B-A3B-Instruct base, retaining the same training recipe: a four-tag scene-graph reasoning format and a dense spatial reward over format, count, accuracy, and grounding.

## Model Description

- **Base Model**: Qwen3-VL-30B-A3B-Instruct (Mixture-of-Experts; ~3B active parameters)
- **Training**: GRPO (Group Relative Policy Optimization) with dense spatial rewards via Thinking Machines' Tinker
- **Training Data**: STVQA-7K (7,587 spatial VQA samples)
- **Authors**: Hunar Batra, Haoqin Tu, Hardy Chen, Yuanze Lin, Cihang Xie, Ronald Clark
- **Institutions**: University of Oxford, UC Santa Cruz

## Key Features

- **Structured Spatial Reasoning**: Constructs question-focused scene subgraphs with objects, bounding boxes, and relations
- **Dense Spatial Rewards**: Multi-objective reward function enforcing format, count, accuracy, and spatial grounding
- **9 Spatial Reasoning Categories**: Relations, reach, size, orientation, instance location, depth, distance, count, and existence
- **MoE Efficiency**: 30B total parameters with only ~3B active per token — comparable quality to dense 30B models at a fraction of the compute

## Inference Template

Same four-tag format as SpatialThinker-7B:

```
You FIRST observe the image in <observe> </observe> tags, then visualise the relevant scene graph in <scene> </scene> tags, followed by thinking about the reasoning process as an internal monologue within <think> </think> tags and then provide the final answer. The final answer MUST BE put within <answer> </answer> tags, and only return the final choice including the correct option and answer within the answer tags, e.g., <answer> (A) cat </answer>.

Image size: {Width} x {Height}
```

## Usage

```python
from transformers import Qwen3VLForConditionalGeneration, AutoProcessor
from PIL import Image

model = Qwen3VLForConditionalGeneration.from_pretrained(
    "hunarbatra/SpatialThinker-30B",
    torch_dtype="auto",
    device_map="auto"
)
processor = AutoProcessor.from_pretrained("hunarbatra/SpatialThinker-30B")

# Load image
image = Image.open("your_image.jpg")
width, height = image.size

# Prepare prompt with template
template = f"""You FIRST observe the image in <observe> </observe> tags, then visualise the relevant scene graph in <scene> </scene> tags, followed by thinking about the reasoning process as an internal monologue within <think> </think> tags and then provide the final answer. The final answer MUST BE put within <answer> </answer> tags, and only return the final choice including the correct option and answer within the answer tags, e.g., <answer> (A) cat </answer>.

Image size: {width} x {height}"""

question = "Where is the cat relative to the couch? (A) on top of (B) in front of (C) behind (D) beside"

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": image},
            {"type": "text", "text": template + "\n\n" + question},
        ],
    }
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=[text], images=[image], return_tensors="pt").to(model.device)

generated_ids = model.generate(**inputs, max_new_tokens=2048)
output = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(output)
```

## Training Details

- **Framework**: Thinking Machines' [Tinker](https://thinkingmachines.ai/tinker/) (LoRA on remote H100 cluster)
- **Steps**: 75
- **Batch size**: 16 prompts × 8 rollouts = 128 generations/step
- **Optimizer**: AdamW, lr=1e-6, KL coefficient=1e-2 (low_var_kl)
- **LoRA**: rank=64 on the language tower

The model was trained with several rollout-side fixes that lift the Qwen3-VL-Instruct base's format-pass rate from ~78% to ~96% during training:
- Forced `<observe>\n` assistant prefix (matches the four-tag schema the model is trained to produce)
- Postprocess rewrites for `<tool_call>` → `<think>` (the Instruct base's tool-use prior occasionally leaks)
- Repairs for orphan/unclosed `<think>` tags

## Citation

```bibtex
@misc{batra2025spatialthinkerreinforcing3dreasoning,
  title={SpatialThinker: Reinforcing 3D Reasoning in Multimodal LLMs via Spatial Rewards},
  author={Hunar Batra and Haoqin Tu and Hardy Chen and Yuanze Lin and Cihang Xie and Ronald Clark},
  year={2025},
  eprint={2511.07403},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2511.07403},
}
```

## Links

- 📄 **Paper**: [arXiv:2511.07403](https://arxiv.org/abs/2511.07403)
- 🌐 **Project Page**: [hunarbatra.com/SpatialThinker](https://hunarbatra.com/SpatialThinker)
- 💻 **GitHub**: [github.com/hunarbatra/SpatialThinker](https://github.com/hunarbatra/SpatialThinker)
- 🤗 **Dataset**: [hunarbatra/STVQA-7K](https://huggingface.co/datasets/hunarbatra/STVQA-7K)
- 🤗 **7B variant**: [hunarbatra/SpatialThinker-7B](https://huggingface.co/hunarbatra/SpatialThinker-7B)