# GPT-6 Luna vs Phi-4

> GPT-6 Luna is the stronger model overall, scoring 53.3 to 31.2 on the Noometry Index. Phi-4 costs 2.3× less per token, which makes it the better buy when GPT-6 Luna's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/gpt-6-luna-vs-phi-4
- Last updated: 2026-10-10
- Shared benchmarks: 21

## Summary

- They share 21 benchmarks with published results for both. GPT-6 Luna scores higher in 9 categories and Phi-4 in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-6 Luna leads 76.1 to 20.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 98.9% for GPT-6 Luna and 13.8% for Phi-4.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $0.10 / $0.50 for GPT-6 Luna.
- GPT-6 Luna accepts more context: 1.05M tokens versus 128K.
- Phi-4 has downloadable open weights; the other is API-only.

## Snapshot

| | GPT-6 Luna | Phi-4 |
|---|---|---|
| Provider | OpenAI | Microsoft |
| Noometry Index | 53.3 | 31.2 |
| Rank | 36 | 279 |
| Context | 1.05M | 128K |
| Input $/M | $0.10 | $0.07 |
| Output $/M | $0.50 | $0.14 |
| Weights | Proprietary | Open |

## Coding

- GPT-6 Luna: 55.5 (#25)
- Phi-4: 34.4 (#239)

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| LMArena Coding | 1439 | 1231 |
| DeepSWE | 66.6% | — |
| FrontierCode | 42.4% | — |
| LMArena WebDev | 1581 | — |
| SciCode | 54.6% | — |
| BigCodeBench Instruct | — | 45.5% |
| LiveBench Coding | — | 30.7% |
| BigCodeBench Complete | — | 55.4% |
| ALE-Bench | 1,577 | — |

## Agentic & Tool Use

- GPT-6 Luna: 33.3 (#54)
- Phi-4: 22.8 (#128)

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| APEX-Agents | 44.3% | — |
| Berkeley Function Calling Leaderboard | — | 28.8% |
| BALROG | — | 11.6% |
| GDP.pdf | 23% | — |

## Reasoning

- GPT-6 Luna: 48.2 (#41)
- Phi-4: 17.7 (#291)

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| Chess Puzzles | 31% | 1% |
| LMArena Hard Prompts | 1411 | 1220 |
| Epoch Capabilities Index | 156.28 | 130.42 |
| ARC-AGI-2 | 59.3% | — |
| NYT Connections (extended) | 68.7% | — |
| ARC-AGI-1 | 86.7% | — |
| CritPt | 19.4% | — |
| LiveBench Reasoning | — | 47.8% |
| Mystery Game Puzzles | 7% | — |
| DTBench | 90.1% | — |
| LiveBench Data Analysis | — | 45.2% |
| LMCA | 44.5% | — |
| LiveBench | — | 41.6% |

## Math

- GPT-6 Luna: 76.1 (#15)
- Phi-4: 20.8 (#285)

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.9% | 13.8% |
| LMArena Math | 1416 | 1246 |
| FrontierMath (Tiers 1-3) | 78.9% | — |
| FrontierMath Tier 4 | 56.1% | — |
| ProofBench | 64% | — |
| LiveBench Math | — | 42% |
| MATH Level 5 | — | 64.9% |

## Knowledge

- GPT-6 Luna: 57.0 (#41)
- Phi-4: 32.6 (#209)

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| GPQA Diamond | 90.5% | 56.1% |
| LMArena Expert | 1444 | 1203 |
| SimpleQA Verified | 41.4% | — |
| Confabulations | — | 29.4% |
| Vectara Hallucination Rate | — | 3.7% |
| MMLU | — | 84.8% |

## Multimodal

- GPT-6 Luna: 42.4 (#30)
- Phi-4: —

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| LMArena Vision | 1217 | — |
| Blueprint-Bench 2 | 31.2% | — |
| Furniture Assembly | 44.2% | — |

## Multilingual

- GPT-6 Luna: 50.5 (#117)
- Phi-4: 37.2 (#237)

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| LMArena Non-English | 1386 | 1197 |
| LMArena Chinese | 1433 | 1212 |
| LMArena French | 1420 | 1224 |
| LMArena German | 1369 | 1222 |
| LMArena Japanese | 1369 | 1158 |
| LMArena Korean | 1360 | 1151 |
| LMArena Russian | 1394 | 1209 |
| LMArena Spanish | 1393 | 1234 |

## Instruction Following

- GPT-6 Luna: 74.3 (#99)
- Phi-4: 60.4 (#251)

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1409 | 1201 |
| LiveBench Instruction Following | — | 58.4% |

## Long Context

- GPT-6 Luna: 43.0 (#111)
- Phi-4: 36.9 (#226)

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1409 | 1217 |

## Writing & Preference

- GPT-6 Luna: 58.3 (#119)
- Phi-4: 40.5 (#244)

| Benchmark | GPT-6 Luna | Phi-4 |
|---|---|---|
| LMArena Text | 1391 | 1217 |
| LMArena Creative Writing | 1363 | 1182 |
| LMArena Multi-Turn | 1396 | 1206 |
| Short-Story Creative Writing | — | 62.6% |
| LiveBench Language | — | 25.6% |

## FAQ

### Is GPT-6 Luna better than Phi-4?

GPT-6 Luna is the stronger model overall, scoring 53.3 to 31.2 on the Noometry Index. Phi-4 costs 2.3× less per token, which makes it the better buy when GPT-6 Luna's lead doesn't matter for your workload.

### Which is cheaper, GPT-6 Luna or Phi-4?

Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; GPT-6 Luna lists at $0.10 and $0.50.

### Is GPT-6 Luna or Phi-4 better for coding?

GPT-6 Luna scores higher on coding benchmarks: 55.5 versus 34.4 in the Noometry coding category.

### Which has the bigger context window?

GPT-6 Luna does, with 1.05M tokens against 128K.

### How many benchmarks do GPT-6 Luna and Phi-4 share?

21 benchmarks have published results for both models. GPT-6 Luna has 42 scored results on Noometry and Phi-4 has 37.