Microsoft
EfficientDEPRECATED

Phi-2

Legacy

Original 2.7B Phi-2 from late 2023. Deprecated; Phi-3.5 mini is the modern equivalent.

Phi-2 is a efficient AI model from Microsoft. It costs $0.0700 per million input tokens and $0.0700 per million output tokens (blended $0.0700/M), with a 2,000-token context window.

RatesUSD per million tokens. Blend is 70% input, 30% output, our calculation.

INPUT
$0.0700/M
per million input tokens
OUTPUT
$0.0700/M
per million output tokens
BLENDED 70/30
$0.0700/M
unchanged since 3 May
CONTEXT
2,000
tokens

Hand-maintained from Microsoft's published pricing; no check date on record.

What it is good at
  • Historic small-model baseline
Typical use cases
  • Legacy Phi-2 fine-tunes

Benchmarks

vs. best public score
MMLU56.7%
Multitask academic knowledge across 57 subjects. Source: Microsoft (self-reported), MMLU, read 2026-11-10.
Python function synthesis from docstrings. Source: Microsoft (self-reported), HumanEval, read 2026-11-10.
Hand-curated from each provider's published reports and public leaderboards. Methodology varies across sources, treat as directional rather than authoritative.

More from Microsoft

See all 8 →

Frequently asked questions

How much does Phi-2 cost?

Phi-2 costs $0.0700 per million input tokens and $0.0700 per million output tokens, for a blended reference rate of $0.0700 per million tokens.

What is Phi-2's context window?

Phi-2 supports up to 2,000 tokens of context in a single request.

What is Phi-2 best for?

Phi-2 is well suited to Historic small-model baseline and Legacy Phi-2 fine-tunes.

Who makes Phi-2?

Phi-2 is developed and served by Microsoft. It was released in Dec 2023.

Terms used on this page