Phi 4
Microsoft · released Jan 10, 2025
[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...
Specification
- Context window
- 16K
- Max output
- 16K
- Knowledge cutoff
- Jun 30, 2024
- Parameters
- 14.7B
- Licence
- mit
- Serving providers
- 1
- Moderated
- No
- Uptime
- 100.0%
Intelligence
4.6
15th percentile
Coding
—
Coding Index
Agentic
—
Agentic Index
Output speed
76 t/s
Median across providers
Latency
304ms
Time to first token
Cost per task
$0.007
Estimated
Benchmarks
Where the score comes from
The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.
Evaluation scores
Percentage correct · higher is better
- MMLU-Pro71.4%
- GPQA Diamond57.5%
- SciCode26.0%
- IFBench23.5%
- LiveCodeBench23.1%
- AIME 202518.0%
- Humanity's Last Exam3.8%
- Terminal-Bench Hard3.8%
- τ²-bench (Telecom)0.0%
- AA-LCR (long context)0.0%
An evaluation missing from this list was not run for this model — it is not a zero.
View as table
| Evaluation | Score |
|---|---|
| MMLU-Pro | 71.4% |
| GPQA Diamond | 57.5% |
| SciCode | 26.0% |
| IFBench | 23.5% |
| LiveCodeBench | 23.1% |
| AIME 2025 | 18.0% |
| Humanity's Last Exam | 3.8% |
| Terminal-Bench Hard | 3.8% |
| τ²-bench (Telecom) | 0.0% |
| AA-LCR (long context) | 0.0% |
Against its peers
Intelligence Index · this model highlighted, nearest peers in grey
- Olmo 3 32B ThinkAllen AI6.1
- Hermes 3 70B InstructNous Research4.8
- Phi 4Microsoft4.6
- Mistral LargeMistral AI4.1
- Mixtral 8x22B InstructMistral AI4.0
- Reka Flash 3Reka AI3.7
- Claude 3 HaikuAnthropic3.5
- GPT-3.5 TurboOpenAI3.2
Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.
View as table
| Model | Intelligence |
|---|---|
| Olmo 3 32B Think | 6.1 |
| Hermes 3 70B Instruct | 4.8 |
| Phi 4 | 4.6 |
| Mistral Large | 4.1 |
| Mixtral 8x22B Instruct | 4.0 |
| Reka Flash 3 | 3.7 |
| Claude 3 Haiku | 3.5 |
| GPT-3.5 Turbo | 3.2 |
Percentile among all indexed models
Pricing
What it costs to run
List prices per million tokens, plus what one representative task works out to.
List price
- Input / 1M tokens
- $0.07
- Output / 1M tokens
- $0.14
- Cached input / 1M
- Not offered
- Blended 3:1
- $0.088
One task, estimated
$0.007
- Input tokens
- 50,000
- Output tokens
- 25,000
- Profile
- Standard
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).
Serving providers
Speed and latency figures are medians across these providers, so a widely-served model reports a blend rather than any single endpoint.