Microsoft, open weights
Phi-4
Phi-4 by Microsoft ranks 279th of 354 ranked models on the Noometry Index as of October 2026, with a score of 31.2. Its strongest category is agentic & tool use, where it ranks 128th. API pricing starts at $0.07 per million input tokens and $0.14 per million output tokens, with a 128K-token context window.
Last verified
Specifications
- Noometry rank
- #279 of 354
- Index score
- 31.2
- Evidence
- Confirmed 37 results
- Provider
Microsoft- Released
- December 11, 2024
- Weights
- Open weights
- Reasoning
- No
- Context window
- 128K
- Max output
- 4K
- Input price
- $0.07 / M
- Output price
- $0.14 / M
- Blended price
- $0.0875 / M
- Output speed
- Not measured
- Value
- #15 of 219
- Knowledge cutoff
- October 2023
- Input
- text
- Hugging Face
- microsoft/phi-4
Category scores
Each category score combines every public result we have in that category.
- Coding 34.4
- Agentic & Tool Use 22.8
- Reasoning 17.7
- Math 20.8
- Knowledge 32.6
- Multilingual 37.2
- Instruction Following 60.4
- Long Context 36.9
- Writing & Preference 40.5
| Category | Score | Rank | Results |
|---|---|---|---|
| Coding | 34.4 | #239 | 4 |
| Agentic & Tool Use | 22.8 | #128 | 2 |
| Reasoning | 17.7 | #291 | 4 |
| Math | 20.8 | #285 | 4 |
| Knowledge | 32.6 | #209 | 4 |
| Multilingual | 37.2 | #237 | 1 |
| Instruction Following | 60.4 | #251 | 2 |
| Long Context | 36.9 | #226 | 1 |
| Writing & Preference | 40.5 | #244 | 5 |
Strengths and weaknesses
Categories where Phi-4 places highest and lowest among the models ranked in each, with its score against that category's median.
Strongest categories
| Category | Score | vs median | Rank |
|---|---|---|---|
| Knowledge | 32.6 | −4.7 | #209 of 314, top 67% |
| Coding | 34.4 | −4.4 | #239 of 340, top 71% |
| Long Context | 36.9 | −4.0 | #226 of 296, top 77% |
Weakest categories
| Category | Score | vs median | Rank |
|---|---|---|---|
| Math | 20.8 | −15.8 | #285 of 327, top 88% |
| Reasoning | 17.7 | −5.9 | #291 of 350, top 84% |
| Agentic & Tool Use | 22.8 | −7.6 | #128 of 154, top 84% |
Closest competitors
The models ranked just above and below Phi-4. When scores are this close, price and speed are often the better way to choose.
| Model | Rank | Score | Blended $/M | Speed | |
|---|---|---|---|---|---|
| Qwen-14B | #275 | 31.4 | — | — | Compare |
| Phi 3 Mini 4k Instruct June 2024 | #276 | 31.3 | — | — | Compare |
| Gemma 1.1 7b IT | #277 | 31.3 | — | — | Compare |
| Mistral Small 3 | #278 | 31.2 | $0.0575 | — | Compare |
| Mistral Small 3.2 | #280 | 31.2 | $0.13 | 68 | Compare |
| Amazon Nova Pro | #281 | 31.0 | $1.40 | — | Compare |
| Llama 4 Maverick | #282 | 30.9 | $0.30 | 456 | Compare |
| Phi-4 Mini | #283 | 30.9 | $0.13 | — | Compare |
Sponsored placements are available on pages like this one. Advertise on Noometry
Benchmark results
Every published result we track, with its source. Bold rows are the ones used for ranking; where several exist we prefer independent runs over self-reported numbers.
Coding
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| BigCodeBench Instruct | 45.5% | #19 of 64, top 30% | BigCodeBench | 2024-12-13 | |
| LiveBench Coding | 30.7% | #31 of 39, top 80% | Epoch AI | ||
| LMArena Coding | 1231 | #230 of 294, top 79% | LMArena | 2026-10-08 | |
| BigCodeBench Complete | 55.4% | #17 of 66, top 26% | BigCodeBench | 2024-12-13 |
Agentic & Tool Use
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| Berkeley Function Calling Leaderboard | 28.8% | #35 of 49, top 72% | prompt | Berkeley Function Calling Leaderboard | |
| BALROG | 11.6% | #32 of 35, top 92% | Epoch AI |
Reasoning
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| Chess Puzzles | 1% | #107 of 129, top 83% | Epoch AI | 2026-08-28 | |
| LiveBench Reasoning | 47.8% | #21 of 39, top 54% | Epoch AI | ||
| LMArena Hard Prompts | 1220 | #231 of 297, top 78% | LMArena | 2026-10-08 | |
| LiveBench Data Analysis | 45.2% | #30 of 39, top 77% | Epoch AI | ||
| Epoch Capabilities Index | 130.42 | #137 of 213, top 65% | Epoch AI | 2024-12-12 | |
| LiveBench | 41.6% | #29 of 39, top 75% | Epoch AI |
Math
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 13.8% | #131 of 173, top 76% | Epoch AI | 2025-02-25 | |
| LiveBench Math | 42% | #25 of 39, top 65% | Epoch AI | ||
| LMArena Math | 1246 | #216 of 285, top 76% | LMArena | 2026-10-08 | |
| MATH Level 5 | 64.9% | #35 of 79, top 45% | Epoch AI | 2025-01-31 |
Knowledge
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| GPQA Diamond | 56.1% | #120 of 186, top 65% | Epoch AI | 2025-01-31 | |
| Confabulations (lower is better) | 29.4% | #46 of 51, top 91% | Lech Mazur benchmarks | ||
| Vectara Hallucination Rate (lower is better) | 3.7% | #3 of 96, top 4% | Vectara Hallucination Leaderboard | ||
| LMArena Expert | 1203 | #220 of 273, top 81% | LMArena | 2026-10-08 | |
| MMLU | 84.8% | #8 of 81, top 10% | Epoch AI |
Multilingual
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LMArena Non-English | 1197 | #237 of 297, top 80% | LMArena | 2026-10-08 | |
| LMArena Chinese | 1212 | #228 of 285, top 80% | LMArena | 2026-10-08 | |
| LMArena French | 1224 | #190 of 223, top 86% | LMArena | 2026-10-08 | |
| LMArena German | 1222 | #185 of 231, top 81% | LMArena | 2026-10-08 | |
| LMArena Japanese | 1158 | #172 of 211, top 82% | LMArena | 2026-10-08 | |
| LMArena Korean | 1151 | #177 of 213, top 84% | LMArena | 2026-10-08 | |
| LMArena Russian | 1209 | #231 of 283, top 82% | LMArena | 2026-10-08 | |
| LMArena Spanish | 1234 | #188 of 226, top 84% | LMArena | 2026-10-08 |
Instruction Following
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LiveBench Instruction Following | 58.4% | #29 of 39, top 75% | Epoch AI | ||
| LMArena Instruction Following | 1201 | #234 of 298, top 79% | LMArena | 2026-10-08 |
Long Context
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LMArena Longer Query | 1217 | #236 of 291, top 82% | LMArena | 2026-10-08 |
Writing & Preference
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LMArena Text | 1217 | #240 of 297, top 81% | LMArena | 2026-10-08 | |
| LMArena Creative Writing | 1182 | #239 of 295, top 82% | LMArena | 2026-10-08 | |
| Short-Story Creative Writing | 62.6% | #36 of 39, top 93% | Epoch AI | ||
| LMArena Multi-Turn | 1206 | #236 of 295, top 80% | LMArena | 2026-10-08 | |
| LiveBench Language | 25.6% | #31 of 39, top 80% | Epoch AI |
API pricing by provider
| Route | Input $/M | Output $/M | Cached input $/M | Checked |
|---|---|---|---|---|
| azure | $0.13 | $0.50 | — | 2026-10-10 |
| openrouter | $0.07 | $0.14 | — | 2026-10-10 |
Compare Phi-4
- Phi-4 vs Mistral Small 3
- Phi-4 vs Mistral Small 3.2
- Phi-4 vs Gemma 1.1 7b IT
- Phi-4 vs Amazon Nova Pro
- Phi-4 vs Phi 3 Mini 4k Instruct June 2024
- Phi-4 vs Llama 4 Maverick
- Phi-4 vs GPT-6 Astra
- Phi-4 vs Claude Fable 5.1
- Phi-4 vs Gemini 3.8 Flash
- Phi-4 vs Kimi K3
- Phi-4 vs Grok 4.6
- Phi-4 vs Qwen3.8 Max
- Phi-4 vs GLM-5.3
- Phi-4 vs Muse Spark 1.3
Other Microsoft models
Frequently asked questions
How good is Phi-4?
Phi-4 by Microsoft ranks 279th of 354 ranked models on the Noometry Index as of October 2026, with a score of 31.2. Its strongest category is agentic & tool use, where it ranks 128th. API pricing starts at $0.07 per million input tokens and $0.14 per million output tokens, with a 128K-token context window.
How much does Phi-4 cost?
Phi-4 costs $0.07 per million input tokens and $0.14 per million output tokens on openrouter.
What is Phi-4's context window?
Phi-4 accepts up to 128K tokens of input and can write up to 4K tokens in one response.
Is Phi-4 open source?
Yes. Phi-4's weights are downloadable from Hugging Face (microsoft/phi-4); check the license for commercial terms.
What are Phi-4's strengths and weaknesses?
Relative to other ranked models, Phi-4 places best in knowledge, coding, long context and lowest in math, reasoning, agentic & tool use.
What is Phi-4 best at?
Its best category is agentic & tool use, where it ranks 128th on Noometry.