Released: 2025-02-15

Phi-4-mini

Ultra-compact 3.8B parameter model featuring a 128K context window engineered for edge devices and high-volume routing.

Intelligence Index
75.2/ 100
Calibrated multi-domain composite
Throughput Speed
195tok/sec
Output streaming throughput
Time To First Token
170ms
Initial chunk server latency
LiveBench Contam-Free
68.4%
September 2026 suite score

Sourced Benchmark Evaluations (61)

Verified performance across authoritative benchmarks with provenance tracking.
BenchmarkStatusScoreTrustDateSource TypeProvenance
ARC-AGI-3Active49.4%
93
2025-03-22independentSource ↗
ARC-AGI-2Nearing Saturation59.3%
75
2025-03-15independentSource ↗
FrontierMathActive22%
98
2025-03-12independentSource ↗
PutnamBenchActive40.8%
90
2025-03-11independentSource ↗
SWE-LancerDeprecated48.3%
20
2025-03-09independentSource ↗
Omni-MATHDeprecated58.1%
10
2025-03-09independentSource ↗
SoohakActive59.7%
98
2025-03-07independentSource ↗
FrontierCodeActive41.2%
98
2025-03-07independentSource ↗
Humanity's Last ExamActive33.4%
93
2025-03-05independentSource ↗
StructFlowBenchNearing Saturation58.8%
75
2025-03-05independentSource ↗
AIMENearing Saturation74.6%
75
2025-03-05independentSource ↗
OlympiadBenchActive64.4%
80
2025-03-04independentSource ↗
MathArenaActive62.8%
99
2025-03-03independentSource ↗
FireBenchActive58.1%
95
2025-03-03independentSource ↗
Terminal-BenchActive49.7%
80
2025-03-03independentSource ↗
OpenAI-MRCRActive63.2%
80
2025-03-02independentSource ↗
SimpleQAActive39.1%
90
2025-03-02independentSource ↗
CodeContestsDeprecated46.2%
10
2025-03-02independentSource ↗
ComplexBenchNearing Saturation56.7%
80
2025-03-01independentSource ↗
SWE-bench ProActive44%
90
2025-03-01independentSource ↗
LongBench-ProActive61.7%
80
2025-03-01independentSource ↗
ZeroSCROLLSNearing Saturation60.2%
78
2025-02-28independentSource ↗
MultiPL-EDeprecated65.3%
10
2025-02-28independentSource ↗
LiveCodeBenchSaturated58.2%
39
2025-02-27independentSource ↗
AgentIFNearing Saturation47%
80
2025-02-27independentSource ↗
MATHSaturated76.9%
25
2025-02-27independentSource ↗
InfiniteBenchNearing Saturation63.9%
65
2025-02-27independentSource ↗
MercuryDeprecated62.5%
10
2025-02-26independentSource ↗
SCROLLSNearing Saturation60.2%
65
2025-02-26independentSource ↗
MMLU-ProNearing Saturation69.2%
75
2025-02-26independentSource ↗
RULERNearing Saturation69.2%
84
2025-02-25independentSource ↗
TabMWPSaturated72.2%
25
2025-02-25independentSource ↗
CyberSecEvalDeprecated63.9%
10
2025-02-25independentSource ↗
SWE-bench VerifiedSaturated55.4%
25
2025-02-24independentSource ↗
MMLU-ReduxSaturated73.7%
25
2025-02-24independentSource ↗
QuALITYNearing Saturation66.2%
78
2025-02-24independentSource ↗
LongBenchNearing Saturation64.7%
65
2025-02-23independentSource ↗
SWE-benchDeprecated48.3%
10
2025-02-23independentSource ↗
MGSMNearing Saturation74.6%
65
2025-02-23independentSource ↗
AGIEvalSaturated65.4%
25
2025-02-22independentSource ↗
QASPERNearing Saturation62.4%
65
2025-02-22independentSource ↗
TruthfulQASaturated66.2%
25
2025-02-22independentSource ↗
GSM-HardDeprecated75.4%
10
2025-02-21independentSource ↗
SuperGLUESaturated77.5%
25
2025-02-21independentSource ↗
CodeXGLUEDeprecated66.7%
23
2025-02-20independentSource ↗
MT-BenchNearing Saturation7.1%
75
2025-02-20independentSource ↗
BIG-Bench HardSaturated74.5%
25
2025-02-20independentSource ↗
WinoGrandeSaturated79%
28
2025-02-19independentSource ↗
MBPP+Nearing Saturation72.4%
65
2025-02-19independentSource ↗
IFEvalDeprecated67.8%
25
2025-02-19independentSource ↗
GPQANearing Saturation66.2%
78
2025-02-19independentSource ↗
Needle-in-a-HaystackSaturated95%
44
2025-02-18independentSource ↗
HumanEval+Saturated76.7%
25
2025-02-18independentSource ↗
CommonsenseQASaturated79.8%
28
2025-02-18independentSource ↗
MBPPSaturated75.3%
25
2025-02-17independentSource ↗
GSM8KSaturated84.8%
25
2025-02-17independentSource ↗
GLUESaturated82.1%
28
2025-02-17independentSource ↗
MMLUSaturated76.7%
25
2025-02-16independentSource ↗
HumanEvalSaturated80.9%
25
2025-02-16independentSource ↗
ARC-AGISaturated71.4%
25
2025-02-16independentSource ↗
LiveBench (Continuous)Active68.4%
94
2025-02-01independentSource ↗
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run phi-4-mini

# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "phi-4-mini",
    "messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
  }'

Evaluation & Sourcing Notes

Scores listed for Phi-4-mini represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.

All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.