math benchmark
PhiBench
PhiBench is an internal benchmark designed to evaluate diverse skills and reasoning abilities of language models, covering a wide range of tasks including coding (debugging, extending incomplete code, explaining code snippets) and mathematics (identifying proof errors, generating related problems). Created by Microsoft's research team to address limitations of standard academic benchmarks and guide the development of the Phi-4 model.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
PhiBench leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | MI | 74.2% | 100.0% | 3 | C | |
| 2 | MI | 70.6% | 50.0% | 3 | C | |
| 3 | MI | 56.2% | 0.0% | 3 | C |
Score distribution
Top published rows on the benchmark's original scale.
PhiBench highlights
The top published results on this benchmark's own scale.
What is PhiBench?
Definition and scoring fields from the benchmark registry.
PhiBench is an internal benchmark designed to evaluate diverse skills and reasoning abilities of language models, covering a wide range of tasks including coding (debugging, extending incomplete code, explaining code snippets) and mathematics (identifying proof errors, generating related problems). Created by Microsoft's research team to address limitations of standard academic benchmarks and guide the development of the Phi-4 model.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- PhiBench
- Modality
- text
- Primary category
- math
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- phibench|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about PhiBench.
Which model scores highest on PhiBench?
Phi 4 Reasoning Plus is currently ranked first with 74.2%.
What does PhiBench measure?
PhiBench is an internal benchmark designed to evaluate diverse skills and reasoning abilities of language models, covering a wide range of tasks including coding (debugging, extending incomplete code, explaining code snippets) and mathematics (identifying proof errors, generating related problems). Created by Microsoft's research team to address limitations of standard academic benchmarks and guide the development of the Phi-4 model.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
3 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.