Phi-4 is a 14 billion parameter dense decoder-only transformer trained on 9.8 trillion tokens, designed for reasoning-intensive tasks where smaller model size is critical. It exceeds its teacher model on STEM-focused benchmarks, achieving 84.8 on MMLU and 80.4 on MATH, and supports a 16,384 token context window. Built with a data-quality-first approach, Phi-4 strategically incorporates synthetic training data rather than relying solely on organic web content or code. It excels at math problem-solving, code generation, and multi-step reasoning, optimized for memory and compute-constrained environments. The model is trained primarily on English text with multilingual capability (about 8% multilingual data), and follows a chat instruction format. It is fully open-weight and available on Hugging Face, making it accessible for commercial and research applications.
Phi-4 runs on 1 route through the Opper gateway. Compare residency, ZDR, and training posture at a glance — full data-handling detail per route below.
| Provider | Region | Zero data retention | Training | Input | Output |
|---|---|---|---|---|---|
| US | Zero data retention | No | $0.07 | $0.14 |
Phi-4 runs on a single route through the Opper gateway today, so its availability is that provider's availability.
Name a second model in the same request and the gateway tries it on retriable errors, so a busy hour never has to reach your users. Set up a fallback chain.
Measured over the last 30 days from each provider's official status feed via StatusGator. Refreshed hourly. See uptime for every provider Opper monitors.
Each route hosting Phi-4 has its own privacy posture, residency, and GDPR terms. Postures are maintained by Opper with a last-verification timestamp.
Zero data retention is on by default on Pay-as-you-go — no action required. No training on customer data. US; unknown; DPA available.
Independent benchmark scores — composite indices for reasoning, coding, and math, plus individual eval scores where available.
Call Phi-4 through the Opper gateway with one API key. Let your coding agent set it up, or call it directly — Opper is drop-in compatible with the OpenAI, Anthropic, and Google AI SDKs.
Copy this and paste it into a coding agent like Claude Code, Cursor or Codex and it'll wire up Opper for you.
import OpenAI from "openai";const client = new OpenAI({apiKey: process.env.OPPER_API_KEY,baseURL: "https://api.opper.ai/v3/compat",});const completion = await client.chat.completions.create({model: "deepinfra/microsoft/phi-4",messages: [{ role: "user", content: "Hello" }],});console.log(completion.choices[0].message.content);
Side-by-side on privacy, EU hosting, pricing, and benchmarks.
