We're excited to announce improvements to our MiMo-VL (MiMo-VL-7B-RL-2508 and MiMo-VL-7B-SFT-2508), featuring enhanced performance across multiple benchmarks, improved thinking control capabilities, and better user experience.
📈 Performance Improvements
MiMo-VL-7B-RL-2508 demonstrates consistent improvements across both image and video benchmarks, achieving notable milestones of 70.6 on MMMU and 70.8 on VideoMME.
A thinking control capability that allows users to turn off the model's reasoning mode using the no_think parameter:
Thinking mode (default behavior): Full reasoning process visible with 100% control success rate;
Non-thinking mode: Direct responses without reasoning, with a 99.84% control success rate
❤️ Enhanced User Experience
Our internal VLM Arena ratings show meaningful improvement in real-world performance:
Current model (MiMo-VL-7B-RL-2508): 1131.2 rating
Previous version (MiMo-VL-7B-RL): 1093.9 rating
These updates deliver a more capable, flexible, and reliable vision-language model for both academic evaluation and practical applications.
😉 Case Study: What are the appealing features of this car?
Xiaomi SU7
MiMo-VL-7B-RL-2508's Answer:
markdown
1<think>...</think>2To determine the appealing features of the Xiaomi SU7, we analyze visual elements and context from the video:
31. Aesthetic Design
4- The car boasts a sleek, aerodynamic exterior with smooth, modern lines.
5- Distinctive lighting (sharp headlights and a continuous, glowing taillight strip) enhances its futuristic and sporty look.
6- Vibrant color and polished finish make it visually striking.
72. Performance & Speed
8- Multiple dynamic shots of the car driving at high speeds on diverse roads (desert, forest, city) emphasize agility and power.
9- Close - ups of spinning wheels and the engine (with lightning-like effects) suggest advanced engineering and high-performance capabilities (likely electric or tech-driven).
103. Versatility
11- The car is shown in varied environments (desert at dusk, forested roads, urban streets), highlighting its adaptability to different terrains and use cases.
124. Technological Innovation
13- The engine close-up with electric-like lightning effects implies cutting-edge technology, positioning the car as a tech-forward vehicle (aligning with Xiaomi’s tech brand identity).
145. Brand Appeal
15Prominent “Xiaomi SU7” branding leverages Xiaomi’s reputation in consumer tech, adding credibility and allure for tech-savvy audiences.
161718In summary, the Xiaomi SU7 appeals through its striking design, high-performance capabilities, versatility, technological innovation, and strong brand association.
Full Evaluation Results
Evaluation Results
Model Recommendation
Both versions of the MiMo-VL-7B-2508 model are now open-sourced on Hugging Face:
Users may perform SFT and RL based on this model. Compared to the previous SFT version, this model demonstrates higher RL stability.
Deployment Parameters
temperature=0.3, topp=0.95
The system prompt is already set in chat_template.json and does not require additional configuration.
Thinking Control
Users can control the thinking mode by appending /no_think to queries:
Thinking mode query (default): "What is the answer to the question in the image?"
Non-thinking mode query: "Identify the text in the image. /no_think"
❗️Important: The /no_think command must be the very last part of user message, which means after /no_think, there shouldn't be any user content like image or video.
Placing Visual Input
For prompts with a single image or video, always place the visual media before the text. For example:
In this report, we share our efforts to build a compact yet powerful VLM, MiMo-VL-7B. MiMo-VL-7B comprises (1) a native resolution ViT encoder that preserves fine-grained visual details, (2) an MLP projector for efficient cross-modal alignment, and (3) our MiMo-7B language model, specifically optimized for complex reasoning tasks.
The development of MiMo-VL-7B involves two sequential training processes: (1) A four-stage pre-training phase, which includes projector warmup, vision-language alignment, general multi-modal pre-training, and long-context Supervised Fine-Tuning (SFT). This phase yields the MiMo-VL-7B-SFT model. (2) A subsequent post-training phase, where we introduce Mixed On-policy Reinforcement Learning (MORL), a novel framework that seamlessly integrates diverse reward signals spanning perception accuracy, visual grounding precision, logical reasoning capabilities, and human/AI preferences. This phase yields the MiMo-VL-7B-RL model.
We open-source MiMo-VL-7B series, including checkpoints of the SFT and RL model.
We believe this report along with the models will provide valuable insights to develop powerful reasoning VLMs that benefit the larger community.
🛤️ During this journey, we find
Incorporating high-quality, broad-coverage reasoning data from the pre-training stage is crucial for enhancing model performance
We curate high-quality reasoning data by identifying diverse queries, employing large reasoning models to regenerate responses with long CoT, and applying rejection sampling to ensure quality.
Rather than treating this as supplementary fine-tuning data, we incorporate substantial volumes of this synthetic reasoning data directly into the later pre-training stages, where extended training yields continued performance improvements without saturation.
Mixed On-policy Reinforcement Learning further enhances model performance, while achieving stable simultaneous improvements remains challenging
We apply RL across diverse capabilities, including reasoning, perception, grounding, and human preference alignment, spanning modalities including text, images, and videos. While this hybrid training approach further unlock model’s potential, interference across data domains remains a challenge.
In general visual-language understanding, MiMo-VL-7B models achieve state-of-the-art open-source results.
Reasoning Tasks
In multi-modal reasoning, both the SFT and RL models significantly outperform all compared open-source baselines across these benchmarks.
[!IMPORTANT]
Results marked with * are obtained using our evaluation framework.
Tasks with ${\dagger}$ are evaluated by GPT-4o.
GUI Tasks
MiMo-VL-7B-RL possess exceptional GUI understanding and grounding capabilities. As a general-purpose VL model, MiMo-VL achieves comparable or even superior performance to GUI-specialized models.
Elo Rating
With our in-house evaluation dataset and GPT-4o judgments, MiMo-VL-7B-RL achieves the highest Elo rating among all evaluated open-source vision-language models, ranking first across models spanning from 7B to 72B parameters.
IV. Deployment
The MiMo-VL-7B series maintain full compatibility with the Qwen2_5_VLForConditionalGeneration architecture for deployment and inference.
