Phi-4-reasoning

Phi-4-reasoning

Flagship-level multimodal open-source model
2025-05-07
LLM
Model capability: thinking
Input:
$1/1M tokens
Output:
$2/1M tokens
Bulk order? Contact your manager for exclusive deals

API Overview

Phi-4 is Microsoft’s flagship multimodal foundation model, positioned as the next-generation AI backbone featuring “powerful reasoning + full-modal fusion.” It supports unified understanding and generation of text, images, and audio.

  • Native Multimodal Architecture: Seamlessly integrates visual, speech, and text inputs, enabling direct code generation from screenshots and summarization extraction from podcasts.
  • Continued Strong Reasoning Capabilities: Inheriting the high-level reasoning performance of the Phi-3 series, it excels in tasks such as mathematics, logic, and coding.
  • Comprehensive Function Calling Support: It can autonomously invoke tools, query databases, and connect to search engines, achieving a closed-loop workflow for real-world tasks.
  • Edge Deployment Optimization: Through quantization and the ONNX GenAI runtime, it can run efficiently on-device on devices such as iPhones, Android phones, and PCs.
  • Well-Established Developer Ecosystem: The accompanying Phi Cookbook provides rich examples covering scenarios such as frontend generation, news broadcasting, and voice interaction.

───────────────────────────────────────────────────────────────────

Core Capabilities

👁️ Unified Understanding of Text, Images, and Audio: Not only can it “describe images,” but it can also integrate speech and text context to perform cross-modal reasoning tasks. 🎙️ End-to-End Speech Integration: Supports audio input parsing and natural speech output, making it easy to build Siri-like voice assistant experiences. 🧩 Native Agent Design: From perception and decision-making to tool invocation, the entire process is completed within the same reasoning pipeline, ensuring more coherent and reliable responses. 📱 True On-Device Multimodal Capability: It can run full-fledged multimodal AI even on resource-constrained devices like mobile phones, without relying on cloud connectivity.

Playground

Log in to explore more features! Click to Log In

API Analytics

API Reference (1)

API DescriptionAPI EndpointRequest MethodStabilityParameter Description
Chat(Talk)
POST
Stable
View Details

API Pricing

$
ModelDescriptionContextOfficial Price302.AI Price

Phi-4-reasoning

-
32000

Input$1 / 1M tokens
Output$2 / 1M tokens

Input$1/ 1M tokens
Output$2/ 1M tokens
Original Price