# GPT-4 vs gpt-oss-20b

> gpt-oss-20b is the stronger model overall, scoring 32.5 to 29.1 on the Noometry Index.

- Canonical page: https://noometry.com/compare/gpt-4-vs-gpt-oss-20b
- Last updated: 2026-10-11
- Shared benchmarks: 24

## Summary

- They share 24 benchmarks with published results for both. GPT-4 scores higher in 1 category and gpt-oss-20b in 7 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in math, where gpt-oss-20b leads 39.4 to 10.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 1.1% for GPT-4 and 65.3% for gpt-oss-20b.
- gpt-oss-20b is cheaper at $0.018 / $0.09 per million input/output tokens, against $30 / $60 for GPT-4.
- gpt-oss-20b accepts more context: 131K tokens versus 8K.
- gpt-oss-20b has downloadable open weights; the other is API-only.

## Snapshot

| | GPT-4 | gpt-oss-20b |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 29.1 | 32.5 |
| Rank | 316 | 255 |
| Context | 8K | 131K |
| Input $/M | $30 | $0.018 |
| Output $/M | $60 | $0.09 |
| Weights | Proprietary | Open |

## Coding

- GPT-4: 31.6 (#283)
- gpt-oss-20b: 37.6 (#192)

| Benchmark | GPT-4 | gpt-oss-20b |
|---|---|---|
| WeirdML | 12.4% | 40.9% |
| LMArena Coding | 1254 | 1306 |
| SciCode | — | 34.4% |
| BigCodeBench Instruct | 46% | — |
| BigCodeBench Complete | 57.2% | — |
| ALE-Bench | — | 566.05 |
| HumanEval+ | 79.3% | — |

## Agentic & Tool Use

- GPT-4: —
- gpt-oss-20b: 9.3 (#154)

| Benchmark | GPT-4 | gpt-oss-20b |
|---|---|---|
| Terminal-Bench | — | 3.4% |
| METR Time Horizons | 36.1% | — |

## Reasoning

- GPT-4: 17.8 (#289)
- gpt-oss-20b: 19.3 (#261)

| Benchmark | GPT-4 | gpt-oss-20b |
|---|---|---|
| Chess Puzzles | 4% | 4% |
| LMArena Hard Prompts | 1241 | 1274 |
| DTBench | 62.7% | 68% |
| LMCA | 17.1% | 14.5% |
| Epoch Capabilities Index | 125.89 | 137.82 |
| Kagi LLM Benchmark | — | 53.2% |
| CritPt | — | 1.4% |
| Mystery Game Puzzles | 12% | — |
| BIG-Bench Hard | 75.1% | — |
| ForecastBench | 57.8 | — |
| HellaSwag | 95.3% | — |
| WinoGrande | 87.5% | — |

## Math

- GPT-4: 10.8 (#309)
- gpt-oss-20b: 39.4 (#103)

| Benchmark | GPT-4 | gpt-oss-20b |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 1.1% | 65.3% |
| LMArena Math | 1269 | 1317 |
| Omni-MATH | — | 56.5% |
| MATH Level 5 | 23% | — |
| GSM8K | 92% | — |

## Knowledge

- GPT-4: 18.4 (#282)
- gpt-oss-20b: 34.6 (#195)

| Benchmark | GPT-4 | gpt-oss-20b |
|---|---|---|
| GPQA Diamond | 35.7% | 60.8% |
| LMArena Expert | 1211 | 1258 |
| MMLU-Pro | — | 74% |
| GPQA (HELM) | — | 59.4% |
| MMLU | 86.4% | — |
| TriviaQA | 84.8% | — |

## Multilingual

- GPT-4: 40.6 (#215)
- gpt-oss-20b: 42.2 (#197)

| Benchmark | GPT-4 | gpt-oss-20b |
|---|---|---|
| LMArena Non-English | 1246 | 1268 |
| LMArena Chinese | 1242 | 1314 |
| LMArena German | 1251 | 1255 |
| LMArena Japanese | 1209 | 1244 |
| LMArena Korean | 1184 | 1236 |
| LMArena Russian | 1251 | 1278 |
| LMArena Spanish | 1261 | 1267 |
| LMArena French | 1283 | — |

## Instruction Following

- GPT-4: 65.3 (#222)
- gpt-oss-20b: 61.8 (#240)

| Benchmark | GPT-4 | gpt-oss-20b |
|---|---|---|
| LMArena Instruction Following | 1241 | 1236 |
| IFEval | — | 73.2% |

## Long Context

- GPT-4: 37.7 (#212)
- gpt-oss-20b: 37.9 (#209)

| Benchmark | GPT-4 | gpt-oss-20b |
|---|---|---|
| LMArena Longer Query | 1244 | 1250 |

## Writing & Preference

- GPT-4: 34.9 (#268)
- gpt-oss-20b: 35.5 (#265)

| Benchmark | GPT-4 | gpt-oss-20b |
|---|---|---|
| LMArena Text | 1263 | 1287 |
| LMArena Creative Writing | 1244 | 1201 |
| EQ-Bench Creative Writing | 752 | 666 |
| LMArena Multi-Turn | 1257 | 1268 |
| WildBench | — | 73.7% |

## FAQ

### Is GPT-4 better than gpt-oss-20b?

gpt-oss-20b is the stronger model overall, scoring 32.5 to 29.1 on the Noometry Index.

### Which is cheaper, GPT-4 or gpt-oss-20b?

gpt-oss-20b is cheaper. It lists at $0.018 per million input tokens and $0.09 per million output tokens; GPT-4 lists at $30 and $60.

### Is GPT-4 or gpt-oss-20b better for coding?

gpt-oss-20b scores higher on coding benchmarks: 37.6 versus 31.6 in the Noometry coding category.

### Which has the bigger context window?

gpt-oss-20b does, with 131K tokens against 8K.

### How many benchmarks do GPT-4 and gpt-oss-20b share?

24 benchmarks have published results for both models. GPT-4 has 38 scored results on Noometry and gpt-oss-20b has 34.