# Gemini 2.5 Flash vs Phi-4

> Gemini 2.5 Flash is the stronger model overall, scoring 39.3 to 31.2 on the Noometry Index. Phi-4 costs 9.7× less per token, which makes it the better buy when Gemini 2.5 Flash's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/gemini-2-5-flash-vs-phi-4
- Last updated: 2026-10-11
- Shared benchmarks: 24

## Summary

- They share 24 benchmarks with published results for both. Gemini 2.5 Flash scores higher in 9 categories and Phi-4 in 0 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where Gemini 2.5 Flash leads 39.9 to 20.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 73.1% for Gemini 2.5 Flash and 13.8% for Phi-4.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $0.30 / $2.50 for Gemini 2.5 Flash.
- Gemini 2.5 Flash accepts more context: 1.05M tokens versus 128K.
- Phi-4 has downloadable open weights; the other is API-only.

## Snapshot

| | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| Provider | Google | Microsoft |
| Noometry Index | 39.3 | 31.2 |
| Rank | 170 | 279 |
| Context | 1.05M | 128K |
| Input $/M | $0.30 | $0.07 |
| Output $/M | $2.50 | $0.14 |
| Weights | Proprietary | Open |

## Coding

- Gemini 2.5 Flash: 35.8 (#220)
- Phi-4: 34.4 (#239)

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| LMArena Coding | 1424 | 1231 |
| SWE-bench Verified (bash only) | 28.7% | — |
| Aider Polyglot | 55.1% | — |
| WeirdML | 41.9% | — |
| BigCodeBench Instruct | — | 45.5% |
| LiveBench Coding | — | 30.7% |
| BigCodeBench Complete | — | 55.4% |
| ALE-Bench | 661.88 | — |

## Agentic & Tool Use

- Gemini 2.5 Flash: 30.8 (#74)
- Phi-4: 22.8 (#128)

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| Berkeley Function Calling Leaderboard | 56.2% | 28.8% |
| BALROG | 33.5% | 11.6% |
| Terminal-Bench | 17.1% | — |
| TheAgentCompany | 41.1% | — |
| Vending-Bench 2 | 548.84 | — |

## Reasoning

- Gemini 2.5 Flash: 18.1 (#286)
- Phi-4: 17.7 (#291)

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| LMArena Hard Prompts | 1422 | 1220 |
| Epoch Capabilities Index | 143.03 | 130.42 |
| ARC-AGI-2 | 2.5% | — |
| SimpleBench | 41.2% | — |
| Kagi LLM Benchmark | 56.8% | — |
| ARC-AGI-1 | 33.3% | — |
| CritPt | 1.1% | — |
| Chess Puzzles | — | 1% |
| EnigmaEval | 2.7% | — |
| LiveBench Reasoning | — | 47.8% |
| DTBench | 76.5% | — |
| LiveBench Data Analysis | — | 45.2% |
| LMCA | 27.5% | — |
| ForecastBench | 60.6 | — |
| LiveBench | — | 41.6% |

## Math

- Gemini 2.5 Flash: 39.9 (#98)
- Phi-4: 20.8 (#285)

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 73.1% | 13.8% |
| LMArena Math | 1415 | 1246 |
| Omni-MATH | 38.5% | — |
| LiveBench Math | — | 42% |
| MATH Level 5 | — | 64.9% |
| FrontierMath (Feb 2025 set) | 4.8% | — |
| FrontierMath Tier 4 (v1) | 4.2% | — |

## Knowledge

- Gemini 2.5 Flash: 36.4 (#168)
- Phi-4: 32.6 (#209)

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| Confabulations | 16.8% | 29.4% |
| Vectara Hallucination Rate | 7.8% | 3.7% |
| LMArena Expert | 1426 | 1203 |
| GPQA Diamond | — | 56.1% |
| Humanity's Last Exam | 12.1% | — |
| MMLU-Pro | 63.9% | — |
| GPQA (HELM) | 39% | — |
| MMLU | — | 84.8% |

## Multimodal

- Gemini 2.5 Flash: 41.8 (#32)
- Phi-4: —

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| LMArena Vision | 1253 | — |
| GeoBench | 76% | — |
| VPCT | 46.2% | — |
| SpatialViz-Bench | 36.9% | — |

## Multilingual

- Gemini 2.5 Flash: 52.3 (#88)
- Phi-4: 37.2 (#237)

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| LMArena Non-English | 1409 | 1197 |
| LMArena Chinese | 1450 | 1212 |
| LMArena French | 1433 | 1224 |
| LMArena German | 1418 | 1222 |
| LMArena Japanese | 1405 | 1158 |
| LMArena Korean | 1385 | 1151 |
| LMArena Russian | 1415 | 1209 |
| LMArena Spanish | 1421 | 1234 |

## Instruction Following

- Gemini 2.5 Flash: 75.7 (#54)
- Phi-4: 60.4 (#251)

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1405 | 1201 |
| LiveBench Instruction Following | — | 58.4% |
| IFEval | 89.8% | — |

## Long Context

- Gemini 2.5 Flash: 47.5 (#17)
- Phi-4: 36.9 (#226)

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1419 | 1217 |
| Fiction.LiveBench | 77.8% | — |

## Writing & Preference

- Gemini 2.5 Flash: 53.8 (#157)
- Phi-4: 40.5 (#244)

| Benchmark | Gemini 2.5 Flash | Phi-4 |
|---|---|---|
| LMArena Text | 1417 | 1217 |
| LMArena Creative Writing | 1400 | 1182 |
| Short-Story Creative Writing | 76.5% | 62.6% |
| LMArena Multi-Turn | 1408 | 1206 |
| EQ-Bench Creative Writing | 1137 | — |
| WildBench | 81.7% | — |
| LiveBench Language | — | 25.6% |

## FAQ

### Is Gemini 2.5 Flash better than Phi-4?

Gemini 2.5 Flash is the stronger model overall, scoring 39.3 to 31.2 on the Noometry Index. Phi-4 costs 9.7× less per token, which makes it the better buy when Gemini 2.5 Flash's lead doesn't matter for your workload.

### Which is cheaper, Gemini 2.5 Flash or Phi-4?

Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; Gemini 2.5 Flash lists at $0.30 and $2.50.

### Is Gemini 2.5 Flash or Phi-4 better for coding?

Gemini 2.5 Flash scores higher on coding benchmarks: 35.8 versus 34.4 in the Noometry coding category.

### Which has the bigger context window?

Gemini 2.5 Flash does, with 1.05M tokens against 128K.

### How many benchmarks do Gemini 2.5 Flash and Phi-4 share?

24 benchmarks have published results for both models. Gemini 2.5 Flash has 54 scored results on Noometry and Phi-4 has 37.