Intended Use

Primary Use Cases

Phi-4-Reasoning-Vision-15B is a multimodal model designed for chat completion with visual understanding, visual question answering, image analysis and classification, and image-to-text generation. It enables users to submit images alongside text prompts to receive detailed descriptions, extract information, answer questions about visual content, and classify images into categories.

Out-of-Scope Use Cases

Our models are not specifically designed or evaluated for all downstream purposes. Developers should consider common limitations of language models as they select use cases, and evaluate and mitigate for accuracy, safety, and fairness before using within a specific downstream use case, particularly for high-risk scenarios. Developers should be aware of and adhere to applicable laws or regulations (including privacy, trade compliance laws, etc.) that are relevant to their use case, including the model’s focus on English. High‑risk domains requiring certified reliability (medical, legal decisions), or fully autonomous execution without human oversight for financial/destructive actions. Require additional verification and product guardrails Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the license the model is released under.

Responsible AI Considerations

Like other language models, the web agent model can potentially behave in ways that are unfair, unreliable, or offensive. Some of the limiting behaviors to be aware of include: Quality of Service: The model is trained primarily on English text. Languages other than English will experience worse performance. English language varieties with less representation in the training data might experience worse performance than standard American English. Model is not intended to support multilingual use. Representation of Harms & Perpetuation of Stereotypes: These models can over- or under-represent groups of people, erase representation of some groups, or reinforce demeaning or negative stereotypes. Despite safety post-training, these limitations may still be present due to differing levels of representation of different groups or prevalence of examples of negative stereotypes in training data that reflect real-world patterns and societal biases. Inappropriate or Offensive Content: These models may produce other types of inappropriate or offensive content, which may make it inappropriate to deploy for sensitive contexts without additional mitigations that are specific to the use case. Information Reliability: Language models can generate nonsensical content or fabricate content that might sound reasonable but is inaccurate or outdated. Developers should apply responsible AI best practices and are responsible for ensuring that a specific use case complies with relevant laws and regulations (e.g. privacy, trade, etc.). Using safety services like Azure AI Content Safety that have advanced guardrails is highly recommended. Important areas for consideration include: Allocation: Models may not be suitable for scenarios that could have consequential impact on legal status or the allocation of resources or life opportunities (ex: housing, employment, credit, etc.) without further assessments and additional debiasing techniques. High-Risk Scenarios: Developers should assess suitability of using models in high-risk scenarios where unfair, unreliable or offensive outputs might be extremely costly or lead to harm. This includes providing advice in sensitive or expert domains where accuracy and reliability are critical (ex: legal or health advice). Additional safeguards should be implemented at the application level according to the deployment context. Misinformation: Models may produce inaccurate information. Developers should follow transparency best practices and inform end-users they are interacting with an AI system. At the application level, developers can build feedback mechanisms and pipelines to ground responses in use-case specific, contextual information, a technique known as Retrieval Augmented Generation (RAG).

• Generation of Harmful Content: Developers should assess outputs for their context and use available safety classifiers or custom solutions appropriate for their use case.
• Misuse: Other forms of misuse such as fraud, spam, or malware production may be possible, and developers should ensure that their applications do not violate applicable laws and regulations.

Training Data

Training Datasets

Sources:

• Data Generation: Bottom-up seed sites and task proposals, where multi-agent solvers and verifiers produce validated trajectories.
• Grounding: Curated datasets for predicting actions and on-screen coordinates.
• UI Understanding: Visual question answering, captioning, and OCR on web pages using screenshots collected through our data generation pipeline.
• Safety & Instruction-Following: Refusal and harm-related datasets, along with instruction-following data to help models decide when to “terminate” or act appropriately.

We've tested on the following benchmarks:

Benchmark	Link
AI2D	lmms-lab/ai2d · Datasets at Hugging Face
HallusionBench	lmms-lab/HallusionBench · Datasets at Hugging Face
MathVerse	AI4Math/MathVerse · Datasets at Hugging Face
MathVision	MathLLMs/MathVision · Datasets at Hugging Face
MathVista	AI4Math/MathVista · Datasets at Hugging Face
MMMU	MMMU/MMMU · Datasets at Hugging Face
MMStar	Lin-Chen/MMStar · Datasets at Hugging Face
ScreenSpot v2	Voxel51/ScreenSpot-v2 · Datasets at Hugging Face
WeMath	We-Math/We-Math · Datasets at Hugging Face
ZEROBench	jonathan-roberts1/zerobench · Datasets at Hugging Face

Benchmark Results

Benchmark	Score
AI2D_TEST	84.8
HallusionBench	64.4
MathVerse_MINI	44.9
MathVision_MINI	36.2
MathVista_MINI	75.2
MMMU_VAL	54.3
MMStar	64.5
ScreenSpot_v2_Desktop	87.1
ScreenSpot_v2_Mobile	88.6
ScreenSpot_v2_Web	88.8
WeMath	50.1
ZEROBench_sub	17.7