# Gemma 4 31B IT vs GPT-5.2

> GPT-5.2 is the stronger model overall, scoring 54.1 to 43.2 on the Noometry Index. Gemma 4 31B IT costs 32× less per token, which makes it the better buy when GPT-5.2's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/gemma-4-31b-it-vs-gpt-5-2
- Last updated: 2026-10-10
- Shared benchmarks: 30

## Summary

- They share 30 benchmarks with published results for both. Gemma 4 31B IT scores higher in 3 categories and GPT-5.2 in 6 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-5.2 leads 50.1 to 26.4.
- The biggest single-benchmark swing is Chess Puzzles: 5% for Gemma 4 31B IT and 49% for GPT-5.2.
- Gemma 4 31B IT is cheaper at $0.09 / $0.34 per million input/output tokens, against $1.75 / $14 for GPT-5.2.
- GPT-5.2 accepts more context: 400K tokens versus 262K.
- Gemma 4 31B IT has downloadable open weights; the other is API-only.

## Snapshot

| | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| Provider | Google | OpenAI |
| Noometry Index | 43.2 | 54.1 |
| Rank | 85 | 31 |
| Context | 262K | 400K |
| Input $/M | $0.09 | $1.75 |
| Output $/M | $0.34 | $14 |
| Weights | Open | Proprietary |

## Coding

- Gemma 4 31B IT: 41.9 (#98)
- GPT-5.2: 52.2 (#32)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| LMArena WebDev | 1364 | 1415 |
| WeirdML | 52.3% | 72.2% |
| LMArena Coding | 1459 | 1447 |
| ALE-Bench | 925.5 | 1,294 |
| SWE-bench Verified | — | 73.8% |
| SWE-bench Verified (bash only) | — | 72.8% |
| SWE-bench Multilingual | — | 66.7% |
| SciCode | 43.4% | — |
| GSO | — | 27.4% |
| AlgoTune | — | 2.05 |

## Agentic & Tool Use

- Gemma 4 31B IT: —
- GPT-5.2: 40.0 (#24)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| Terminal-Bench | — | 64.9% |
| Berkeley Function Calling Leaderboard | — | 55.9% |
| GDPval | — | 49.7% |
| Remote Labor Index | — | 2.5% |
| τ²-bench Airline | — | 83% |
| τ²-bench Banking | — | 32.2% |
| τ²-bench Retail | — | 81.6% |
| τ²-bench Telecom | — | 89.7% |
| DeepResearch Bench | — | 41.1% |
| LMArena Search | — | 1207 |
| METR Time Horizons | — | 75.3% |
| Vending-Bench 2 | — | 3,591 |

## Reasoning

- Gemma 4 31B IT: 26.4 (#119)
- GPT-5.2: 50.1 (#33)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| Kagi LLM Benchmark | 63.5% | 73.3% |
| NYT Connections (extended) | 70.6% | 83.6% |
| Chess Puzzles | 5% | 49% |
| LMArena Hard Prompts | 1448 | 1445 |
| DTBench | 82.7% | 90.9% |
| LMCA | 39.3% | 43.9% |
| Epoch Capabilities Index | 142.74 | 153.45 |
| ARC-AGI-2 | — | 52.9% |
| SimpleBench | — | 45.8% |
| ARC-AGI-1 | — | 86.2% |
| CritPt | 1.4% | — |
| EnigmaEval | — | 10.4% |
| Thematic Generalization | 53% | — |
| EBR-Bench | — | 23% |
| Mystery Game Puzzles | — | 23% |
| Surface Evolver Bench | 30.6% | — |
| ForecastBench | — | 60.1 |

## Math

- Gemma 4 31B IT: 41.3 (#72)
- GPT-5.2: 59.8 (#32)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 73.3% | 96.1% |
| LMArena Math | 1464 | 1439 |
| FrontierMath (Tiers 1-3) | — | 67.4% |
| FrontierMath Tier 4 | — | 31.7% |
| MathArena Final-Answer Competitions | — | 72% |
| ProofBench | — | 15% |
| FrontierMath (Feb 2025 set) | — | 40.7% |
| FrontierMath Tier 4 (v1) | — | 18.8% |

## Knowledge

- Gemma 4 31B IT: 37.9 (#141)
- GPT-5.2: 59.2 (#29)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| GPQA Diamond | 75.8% | 91.4% |
| SimpleQA Verified | 10.4% | 37.1% |
| Vectara Hallucination Rate | 7.4% | 8.4% |
| LMArena Expert | 1465 | 1445 |
| Humanity's Last Exam | — | 27.8% |

## Multimodal

- Gemma 4 31B IT: 41.0 (#32)
- GPT-5.2: 50.9 (#7)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| LMArena Vision | 1278 | 1268 |
| LMArena Document | 1425 | 1405 |
| VPCT | — | 84% |
| Furniture Assembly | — | 38.3% |

## Multilingual

- Gemma 4 31B IT: 54.0 (#52)
- GPT-5.2: 53.6 (#63)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| LMArena Non-English | 1431 | 1425 |
| LMArena Chinese | 1478 | 1460 |
| LMArena French | 1436 | 1456 |
| LMArena Russian | 1462 | 1440 |
| LMArena Spanish | 1443 | 1433 |
| LMArena German | — | 1448 |
| LMArena Japanese | — | 1421 |
| LMArena Korean | — | 1390 |

## Instruction Following

- Gemma 4 31B IT: 75.6 (#58)
- GPT-5.2: 74.9 (#83)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| LMArena Instruction Following | 1433 | 1417 |

## Long Context

- Gemma 4 31B IT: 44.2 (#67)
- GPT-5.2: 44.0 (#74)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| LMArena Longer Query | 1446 | 1428 |
| CL-bench | — | 18.2% |

## Writing & Preference

- Gemma 4 31B IT: 60.7 (#92)
- GPT-5.2: 67.0 (#30)

| Benchmark | Gemma 4 31B IT | GPT-5.2 |
|---|---|---|
| LMArena Text | 1443 | 1439 |
| LMArena Creative Writing | 1416 | 1402 |
| EQ-Bench Creative Writing | 1368 | 1703 |
| LMArena Multi-Turn | 1452 | 1458 |
| EQ-Bench 4 | 1120 | — |

## FAQ

### Is Gemma 4 31B IT better than GPT-5.2?

GPT-5.2 is the stronger model overall, scoring 54.1 to 43.2 on the Noometry Index. Gemma 4 31B IT costs 32× less per token, which makes it the better buy when GPT-5.2's lead doesn't matter for your workload.

### Which is cheaper, Gemma 4 31B IT or GPT-5.2?

Gemma 4 31B IT is cheaper. It lists at $0.09 per million input tokens and $0.34 per million output tokens; GPT-5.2 lists at $1.75 and $14.

### Is Gemma 4 31B IT or GPT-5.2 better for coding?

GPT-5.2 scores higher on coding benchmarks: 52.2 versus 41.9 in the Noometry coding category.

### Which has the bigger context window?

GPT-5.2 does, with 400K tokens against 262K.

### How many benchmarks do Gemma 4 31B IT and GPT-5.2 share?

30 benchmarks have published results for both models. Gemma 4 31B IT has 35 scored results on Noometry and GPT-5.2 has 67.