ZMIME
Microsoft · Released Dec 11, 2024

Phi-4-mini

Compact Microsoft instruction model tuned for efficient coding assistance, reasoning, and low-latency agent tasks.

  • Open weights
  • Tool calling
Capability
—No ECI score yet
Input
$0.075per 1M tokens
Output
$0.30per 1M tokens
Context
128K4K max output
Pricing

What it costs

Input
$0.075per million tokens
Output
$0.30per million tokens
Blended (3:1)
$0.131cheaper than 88% of priced models

Typical monthly bills

Monthly usageEstimated cost
Side project2M input + 0.5M output tokens$0.30
Team assistant25M input + 5M output tokens$3.38
Production app250M input + 50M output tokens$33.75

Official Azure API price, as listed on models.dev.

Capability

Independent benchmarks

Epoch AI has not published a Capabilities Index score for Phi-4-mini yet.

Specifications

The details

Lab
Microsoft
Released
Dec 11, 2024
Knowledge cutoff
Oct 2023
Context window
128,000 tokens
Max output
4,096 tokens
Inputs
Text
Output
Text
Reasoning
No
Tool calling
Yes
Structured output
No
Weights
Open weights
API model ID
phi-4-minion Azure
Availability
1 API providerlisted on models.dev

Azure model documentation

Lab-reported

What Microsoft claims

Published by the lab at launch. Settings vary, so compare these only with care.

BenchmarkScoreSettingSource
MMLU67.3—Source
Same lab

More from Microsoft

Questions

About Phi-4-mini

How much does Phi-4-mini cost?

Phi-4-mini costs $0.075 input / $0.30 output per million tokens (official Azure API price). At a 3:1 input-to-output mix that is $0.131 per million tokens, cheaper than 88% of the 360 priced models we track.

What is the context window of Phi-4-mini?

Phi-4-mini accepts up to 128,000 tokens per request and can write up to 4,096 tokens in one response.

How good is Phi-4-mini?

Epoch AI has not published a Capabilities Index score for Phi-4-mini yet.

Is Phi-4-mini open source?

Yes. Microsoft publishes the weights, so it can be downloaded and self-hosted.

What inputs does Phi-4-mini support?

Phi-4-mini accepts text and replies in text. It is not a dedicated reasoning model, supports tool calling.

When was Phi-4-mini released?

Microsoft released Phi-4-mini on Dec 11, 2024. Its training data runs to Oct 2023.

What are the best alternatives to Phi-4-mini?

The closest current models from other labs on capability, price and release date are Nova Lite (Amazon, $0.06 / $0.24), Llama-3.2-3B (Meta, $0.10 / $0.335), Pixtral 12B (Mistral AI, $0.15 / $0.15) and Qwen-MT Turbo (Alibaba (Qwen), $0.16 / $0.49).