About
[Microsoft Research](/microsoft) Phi-4 is designed to perform well in complex reasoning tasks and can operate efficiently in situations with limited memory or where quick responses are needed. At 14 billion...
Related Models
Pricing
Input
$0.07
per 1M tokens
Output
$0.14
per 1M tokens
Blended
$0.09
per 1M tokens
Cheaper than 85% of models. Median price is $0.54/1M tokens.
Cost Calculator
Tokens per day1M
100K100M
Daily
$0.09
Monthly
$2.63
vs. Similar Models
LFM2 24B A2BQ:0.0
$0.05-41%
Qwen3.5 0.8BQ:+0.1
$0.02-77%
Jamba 1.6 LargeQ:+0.1
$3.50+3900%
Jamba 1.5 LargeQ:+0.2
$3.50+3900%
Performance
40
tokens/sec
Faster than 7% of models
0.49
seconds
Faster than 89% of models
0.49
seconds
Faster than 93% of models
Market Median
94 tok/s
58% slower
Median TTFT
1.11s
56% faster
Throughput/Dollar
452
tok/s per $/1M
Speed Comparison
Qwen3.5 4B (Non-reasoning)
40 tok/s+0%
Hermes 4 - Llama-3.1 405B (Reasoning)
40 tok/s+1%
Claude 4.1 Opus (Reasoning)
40 tok/s+1%
Context Window
16K
tokens
Larger than 5% of models
Max Output
16K
tokens
100% of context
Benchmarks
MMLU-Pro
71.4%
GPQA Diamond
57.5%
HLE
4.1%
LiveCodeBench
23.1%
SciCode
26.0%
TerminalBench Hard
3.8%
MATH-500
81.0%
AIME
14.3%
AIME 2025
18.0%
IFBench
23.5%
Long Context Recall
0.0%
Tau2
0.0%
Market AverageTop Score