Inference
EN
Microsoft

Phi-4

Phi-4 is a standard model from Microsoft. 12 benchmarks count toward its index, in 7 categories.

  • Open weights
  • Standard
  • Input: text
  • Released 2024-12
RankedEnough results for an index.
IndexAll categories, out of 100.22.4±5.7Median 51.0
CoverageIndex weight that has results.90%
SpeedOutput tokens per second.93/sMedian 67/s
First tokenSeconds to the first output token.0.3 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.07Median $0.60
Output / 1MUS dollars per 1M output tokens.$0.14Median $2.55
ContextMaximum tokens in one request.16KMedian 262K
EloLMArena rating and rank.1217±5 #319Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
microsoft/phi-4
Max outputMaximum output tokens per request.
15K
Knowledge cutoffLast date of training data.
30 Jun 2024
ToolsTool calls on OpenRouter.
No
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
—
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Run locally

14.66B parameters
16-bit
31.8 GB
8-bit
17.4 GB
4-bit
10.3 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy fits in 12 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.1 counted5.0—
CodingCode writing and repair.1 counted23.4—
ReasoningLogic problems and puzzles.2 counted22.1222 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.3 counted29.9239 / 268
MultilingualTasks in many languages.2 counted24.09 / 9
InstructionTasks with strict prompt rules.1 counted11.0—
MathMath problems.2 counted23.7154 / 168

Results

12 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
MATH level 5Math64.9% ±1.137 / 88
Multilingual Grade School MathMultilingual80.6%7 / 13
GPQA diamond · Epoch AIKnowledge56.1% ±2.6132 / 217
Evaluating Large Language Models Trained on CodeCoding82.6%14 / 22
Massive Multitask Language UnderstandingKnowledge84.8%22 / 34
Critical Physics TasksReasoning0.0%150 / 214
OTIS Mock AIME 2024-2025Math13.8% ±3.7152 / 200
Artificial Analysis GPQA DiamondKnowledge57.5%163 / 199
Graduate-Level Google-Proof Q&AKnowledge56.1%117 / 141
Artificial Analysis Humanity's Last ExamKnowledge3.8%202 / 222
Artificial Analysis Long Context ReasoningReasoning0.0%204 / 217
Artificial Analysis IFBenchInstruction23.5%142 / 146
τ²-Bench Tool-Agent-User EvaluationAgentic0.0%148 / 149
MMLU-ProXMultilingual49.9%17 / 17

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BY

History

5 releases
Index23.3 → 22.4
23 Oct 202612 Nov 2026
Output / 1M$0.14 → $0.14
23 Oct 202612 Nov 2026

Questions

How much does Phi-4 cost?
$0.07 per 1M input tokens, $0.14 per 1M output tokens.
How good is Phi-4?
Its index is 22.4 out of 100. That is place 275 of 283 ranked models.
What is Phi-4 best at?
Knowledge: 29.9 out of 100, place 239 of 268.
How fast is Phi-4?
93 output tokens per second, 0.3 s to the first token.
What is the context window of Phi-4?
16K tokens.
Is Phi-4 open source?
Yes. Its weights are published.
When was Phi-4 released?
December 2024.

Badge

Phi-4 on Inference360: place 275 of 283, index 22.4
HTML
<a href="https://inference360.ai/models/phi-4"><img src="https://inference360.ai/badges/phi-4.svg" alt="Phi-4 on Inference360: place 275 of 283, index 22.4" height="20"></a>
Markdown
[![Phi-4 on Inference360: place 275 of 283, index 22.4](https://inference360.ai/badges/phi-4.svg)](https://inference360.ai/models/phi-4)

Compare with

Claude 3 Haiku22.0Gemma 4 E2B21.9Claude 3 Opus23.3Phi-4-reasoning27.1

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeNemotron 3 Ultra50.8 · Free

More from Microsoft

All models
Phi-4 Multimodal InstructUnrankedPhi-3.5-miniUnrankedPhi-4-reasoning27.1Phi-4-reasoning-plusUnrankedWizardLM-2 8x22BUnranked