LLM Reference

Phi-4 Reasoning Vision 15B

Released
2026-03-12
Last refreshed
2026-06-04
Status
Researched 28d ago
Open sourceCommercial use: permittedMultimodalVision

Phi-4 Reasoning Vision 15B is a released vision model with open-source; evaluate it while provider pricing coverage matures.

Use it for

  • Teams evaluating vision

Do not use it for

  • Cost-sensitive launches that need sourced token pricing
  • Strict JSON or tool-calling flows
  • Teams that need a tracked hosted API route today
Specifications
Family
Phi-4
Released
2026-03-12
Parameters
15B
Knowledge cutoff
2025-03
Openness
Open source
License
MITOSI-approvedCommercial use: permitted
Created by

Advancing the state-of-the-art in AI and computing.

Redmond, Washington, United States
Founded 1991
Website
Pricing

No tracked provider token pricing is available yet.

About

15B parameter open-weight multimodal reasoning model from Microsoft Research for vision-language tasks. Supports image captioning, math/science reasoning, and UI understanding. Released March 2026.

Phi-4 Reasoning Vision 15B is an open-source model in the Phi-4 family. The structured metadata tracks multimodal input. Headline tracked benchmarks include MATH-500 75.2 and Massive Multi-discipline Multimodal Understanding 54.3.

Top use-case fit

Vision

1 relevant benchmark in the decision map.

Provider price ladder

No tracked provider token pricing is available for this model yet.

Capabilities

VisionMultimodal

Benchmark peer barsfor Vision

Benchmark scores(2)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
BenchmarkScoreVersionSource
MATH-50075.2MATH-500 (accuracy)https://www.microsoft.com/en-us/research/blog/phi-4-reasoning-vision-and-the-lessons-of-training-a-multimodal-reasoning-model/
Massive Multi-discipline Multimodal Understanding54.3MMMU (accuracy)https://www.microsoft.com/en-us/research/blog/phi-4-reasoning-vision-and-the-lessons-of-training-a-multimodal-reasoning-model/

Migration checks

No linked migration route is available for this model yet.

Compare Phi-4 Reasoning Vision 15B with other models

Show all 32 popular comparisonssorted by 7-day search impressions
Phi-4 Reasoning Vision 15B vs DeepSeek V4 Pro5Phi-4 Reasoning Vision 15B vs Gemini 2.5 Pro5Phi-4 Reasoning Vision 15B vs Mistral Medium 35Phi-4 Reasoning Vision 15B vs Xiaomi MiMo-V2.5-Pro5Phi-4 Reasoning Vision 15B vs GPT-5.45Phi-4 Reasoning Vision 15B vs Gemini 3.1 Flash-Lite5Phi-4 Reasoning Vision 15B vs Qwen3.6-27B4Phi-4 Reasoning Vision 15B vs GPT-2 Large4Phi-4 Reasoning Vision 15B vs Doubao4Phi-4 Reasoning Vision 15B vs ELYZA Japanese Llama 2 7B4Phi-4 Reasoning Vision 15B vs Gemini 3 Pro4Phi-4 Reasoning Vision 15B vs GPT-4 Turbo3Phi-4 Reasoning Vision 15B vs Code Cushman 0023Phi-4 Reasoning Vision 15B vs GPT-2 Medium3Phi-4 Reasoning Vision 15B vs GPT-5.4-Cyber3Phi-4 Reasoning Vision 15B vs Gemini Deep Research Preview3Phi-4 Reasoning Vision 15B vs GPT-5.4 Pro3Phi-4 Reasoning Vision 15B vs Mistral Medium 3 Instruct2Phi-4 Reasoning Vision 15B vs Qwen3.6 Max Preview2Phi-4 Reasoning Vision 15B vs Doubao Pro 256K2Phi-4 Reasoning Vision 15B vs GPT-22Phi-4 Reasoning Vision 15B vs Magistral Small 25061Phi-4 Reasoning Vision 15B vs ELYZA Japanese Llama 2 13B1Phi-4 Reasoning Vision 15B vs Mistral Magistral Small 25091Phi-4 Reasoning Vision 15B vs Gemini 3.1 Pro Preview1Phi-4 Reasoning Vision 15B vs GPT-4o (08-06)1Phi-4 Reasoning Vision 15B vs GPT-2 XL1Phi-4 Reasoning Vision 15B vs Xiaomi MiMo-V2.51Phi-4 Reasoning Vision 15B vs Code Davinci 0011Phi-4 Reasoning Vision 15B vs DeepSeek V3 Base1Phi-4 Reasoning Vision 15B vs GPT-5.2 Codex1Phi-4 Reasoning Vision 15B vs Kimi K2 Thinking Turbo1

Frequently asked questions

When was Phi-4 Reasoning Vision 15B released?

Phi-4 Reasoning Vision 15B was released on 2026-03-12.

What benchmarks has Phi-4 Reasoning Vision 15B been tested on?

Phi-4 Reasoning Vision 15B has been evaluated on 2 benchmarks, including MATH-500, Massive Multi-discipline Multimodal Understanding.