# gpt-oss-20b vs Llama 3.1-70B

> gpt-oss-20b is the stronger model overall, scoring 32.5 to 29.6 on the Noometry Index.

- Canonical page: https://noometry.com/compare/gpt-oss-20b-vs-llama-3-1-70b
- Last updated: 2026-10-12
- Shared benchmarks: 28

## Summary

- They share 28 benchmarks with published results for both. gpt-oss-20b scores higher in 6 categories and Llama 3.1-70B in 3 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where gpt-oss-20b leads 39.4 to 13.5.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 65.3% for gpt-oss-20b and 3.6% for Llama 3.1-70B.
- gpt-oss-20b is cheaper at $0.018 / $0.09 per million input/output tokens, against $0.40 / $0.40 for Llama 3.1-70B.
- gpt-oss-20b accepts more context: 131K tokens versus 128K.

## Snapshot

| | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| Provider | OpenAI | Meta |
| Noometry Index | 32.5 | 29.6 |
| Rank | 255 | 308 |
| Context | 131K | 128K |
| Input $/M | $0.018 | $0.40 |
| Output $/M | $0.09 | $0.40 |
| Weights | Open | Open |

## Coding

- gpt-oss-20b: 37.6 (#192)
- Llama 3.1-70B: 30.3 (#296)

| Benchmark | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| WeirdML | 40.9% | 9% |
| LMArena Coding | 1306 | 1260 |
| SciCode | 34.4% | — |
| BigCodeBench Instruct | — | 46.1% |
| BigCodeBench Complete | — | 54.8% |
| ALE-Bench | 566.05 | — |

## Agentic & Tool Use

- gpt-oss-20b: 9.3 (#154)
- Llama 3.1-70B: 25.1 (#112)

| Benchmark | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| Terminal-Bench | 3.4% | — |
| TheAgentCompany | — | 6.9% |
| BALROG | — | 27.9% |

## Reasoning

- gpt-oss-20b: 19.3 (#261)
- Llama 3.1-70B: 21.6 (#220)

| Benchmark | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| LMArena Hard Prompts | 1274 | 1241 |
| DTBench | 68% | 60% |
| LMCA | 14.5% | 14.8% |
| Epoch Capabilities Index | 137.82 | 125.92 |
| Kagi LLM Benchmark | 53.2% | — |
| CritPt | 1.4% | — |
| Chess Puzzles | 4% | — |

## Math

- gpt-oss-20b: 39.4 (#103)
- Llama 3.1-70B: 13.5 (#304)

| Benchmark | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 65.3% | 3.6% |
| Omni-MATH | 56.5% | 21% |
| LMArena Math | 1317 | 1252 |
| MATH Level 5 | — | 36.7% |

## Knowledge

- gpt-oss-20b: 34.6 (#195)
- Llama 3.1-70B: 24.2 (#269)

| Benchmark | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| GPQA Diamond | 60.8% | 44.2% |
| MMLU-Pro | 74% | 65.3% |
| GPQA (HELM) | 59.4% | 42.6% |
| LMArena Expert | 1258 | 1209 |
| MMLU | — | 80.1% |

## Multilingual

- gpt-oss-20b: 42.2 (#197)
- Llama 3.1-70B: 38.8 (#225)

| Benchmark | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| LMArena Non-English | 1268 | 1219 |
| LMArena Chinese | 1314 | 1215 |
| LMArena German | 1255 | 1222 |
| LMArena Japanese | 1244 | 1132 |
| LMArena Korean | 1236 | 1140 |
| LMArena Russian | 1278 | 1234 |
| LMArena Spanish | 1267 | 1253 |
| LMArena French | — | 1261 |

## Instruction Following

- gpt-oss-20b: 61.8 (#240)
- Llama 3.1-70B: 65.3 (#223)

| Benchmark | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| IFEval | 73.2% | 82.1% |
| LMArena Instruction Following | 1236 | 1231 |

## Long Context

- gpt-oss-20b: 37.9 (#209)
- Llama 3.1-70B: 37.6 (#214)

| Benchmark | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| LMArena Longer Query | 1250 | 1241 |

## Writing & Preference

- gpt-oss-20b: 35.5 (#265)
- Llama 3.1-70B: 35.4 (#267)

| Benchmark | gpt-oss-20b | Llama 3.1-70B |
|---|---|---|
| LMArena Text | 1287 | 1261 |
| LMArena Creative Writing | 1201 | 1232 |
| EQ-Bench Creative Writing | 666 | 784 |
| WildBench | 73.7% | 75.8% |
| LMArena Multi-Turn | 1268 | 1256 |

## FAQ

### Is gpt-oss-20b better than Llama 3.1-70B?

gpt-oss-20b is the stronger model overall, scoring 32.5 to 29.6 on the Noometry Index.

### Which is cheaper, gpt-oss-20b or Llama 3.1-70B?

gpt-oss-20b is cheaper. It lists at $0.018 per million input tokens and $0.09 per million output tokens; Llama 3.1-70B lists at $0.40 and $0.40.

### Is gpt-oss-20b or Llama 3.1-70B better for coding?

gpt-oss-20b scores higher on coding benchmarks: 37.6 versus 30.3 in the Noometry coding category.

### Which has the bigger context window?

gpt-oss-20b does, with 131K tokens against 128K.

### How many benchmarks do gpt-oss-20b and Llama 3.1-70B share?

28 benchmarks have published results for both models. gpt-oss-20b has 34 scored results on Noometry and Llama 3.1-70B has 35.