Model Profile
phi-4
Use this page to decide where this model is a strong fit. Rankings below are benchmark-backed by use case, with explicit confidence and contributor metrics.
Identity
ID: microsoft/phi-4
Author: microsoft
Origin: huggingface_catalog
Arch: unknown
Benchmark Coverage
Scored use cases: 12
Avg confidence: 19.0%
Evidence points: 90
Raw rows: 114
Weighted rows: 24
Catalog Metadata
Parameters: unknown
Context window: 4096
Downloads: 572,008
Intelligence Profile
Dimension Breakdown
No iq benchmarks found
No eq benchmarks found
No accuracy benchmarks found
* Low confidence — limited benchmark evidence for this dimension
2/5 dimensions scored · Last updated Apr 21, 2026
Benchmark Signals
Click through to the benchmark source behind this model profile.
DuckDB NSQL Leaderboard
all_execution_accuracy
Normalized value 65.4% · confidence 100.0%
Strongest impact in Metric definition workshop
duckdb_nsql_leaderboard.all_execution_accuracy · Apr 1, 2026
Vectara HHEM Leaderboard
overall_hallucination_error_pct
Normalized value 91.2% · confidence 100.0%
Strongest impact in Meeting Summarization
vectara_hhem_leaderboard.overall_hallucination_error_pct · Apr 1, 2026
LanguageBench
overall:mean
Normalized value 59.6% · confidence 100.0%
Strongest impact in Archaic and historical translation
languagebench.overall_mean · Apr 1, 2026
LanguageBench Grammar/Clarity Official (Split)
grammar_clarity_score_pct
Normalized value 66.0% · confidence 100.0%
Strongest impact in Text tagging and routing
languagebench_grammar_clarity_official.grammar_clarity_score_pct · Apr 1, 2026
LanguageBench Translation Official (Split)
translation_to:bleu
Normalized value 23.0% · confidence 100.0%
Strongest impact in Archaic and historical translation
languagebench_translation_official.translation_to_bleu · Apr 1, 2026
Vectara HHEM Leaderboard
law_hallucination_error_pct
Normalized value 91.5% · confidence 100.0%
Strongest impact in Legal translation
vectara_hhem_leaderboard.law_hallucination_error_pct · Apr 1, 2026
Some fit rows have limited benchmark evidence.
11 of 12 scored use cases have low confidence or thin contributor coverage.
Coverage Diagnostics
actively scoredUse-Case Scores
65
Total Measurements
114
Weighted Measurements
24
Weighted Sources
6
Raw Source Coverage
Weighted Source Coverage
Best Use Cases for This Model
| Use Case | Score |
|---|---|
| Metric definition workshop use_case.data.metric_definition_workshop | 11.6% |
| Legal translation use_case.legal.legal_translation | 11.2% |
| Archaic and historical translation use_case.history.archaic_translation | 10.9% |
| Multilingual Customer Support use_case.cx.multilingual_support | 10.4% |
| Translation and localization use_case.business.translation_localization | 10.4% |
| Text-to-SQL analyst assistant use_case.data.text_to_sql | 10.4% |
| Data quality assistant use_case.data.data_quality_assistant | 10.2% |
| Political risk brief use_case.geo.political_risk_brief | 9.9% |
| SQL debugging use_case.data.sql_debugging | 9.7% |
| Cross-lingual summary use_case.business.cross_lingual_summary | 9.7% |
| Meeting Summarization use_case.business.meeting_summarization | 9.6% |
| Text tagging and routing use_case.business.text_tagging | 9.2% |