# Claude Sonnet 4 vs gpt-oss-20b

> Claude Sonnet 4 is the stronger model overall, scoring 40.6 to 32.1 on the Noometry Index. gpt-oss-20b costs 167× less per token, which makes it the better buy when Claude Sonnet 4's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/claude-sonnet-4-vs-gpt-oss-20b
- Last updated: 2026-10-10
- Shared benchmarks: 32

## Summary

- They share 32 benchmarks with published results for both. Claude Sonnet 4 scores higher in 8 categories and gpt-oss-20b in 1 category; 9 gaps are clear of the uncertainty.
- The widest gap is in agentic & tool use, where Claude Sonnet 4 leads 38.3 to 9.2.
- The biggest single-benchmark swing is Kagi LLM Benchmark: 73% for Claude Sonnet 4 and 53.2% for gpt-oss-20b.
- gpt-oss-20b is cheaper at $0.018 / $0.09 per million input/output tokens, against $3 / $15 for Claude Sonnet 4.
- Claude Sonnet 4 accepts more context: 200K tokens versus 131K.
- gpt-oss-20b has downloadable open weights; the other is API-only.

## Snapshot

| | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| Provider | Anthropic | OpenAI |
| Noometry Index | 40.6 | 32.1 |
| Rank | 139 | 249 |
| Context | 200K | 131K |
| Input $/M | $3 | $0.018 |
| Output $/M | $15 | $0.09 |
| Weights | Proprietary | Open |

## Coding

- Claude Sonnet 4: 44.0 (#70)
- gpt-oss-20b: 37.1 (#185)

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| SciCode | 40% | 34.4% |
| WeirdML | 46.1% | 40.9% |
| LMArena Coding | 1414 | 1307 |
| ALE-Bench | 655.35 | 566.05 |
| SWE-bench Verified (bash only) | 64.9% | — |
| Aider Polyglot | 61.3% | — |
| GSO | 4.9% | — |

## Agentic & Tool Use

- Claude Sonnet 4: 38.3 (#29)
- gpt-oss-20b: 9.2 (#146)

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| Terminal-Bench | — | 3.4% |
| TheAgentCompany | 33.1% | — |
| Cybench | 35% | — |
| DeepResearch Bench | 46.6% | — |
| OSWorld | 43.9% | — |
| METR Time Horizons | 62% | — |

## Reasoning

- Claude Sonnet 4: 23.0 (#169)
- gpt-oss-20b: 18.8 (#251)

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| Kagi LLM Benchmark | 73% | 53.2% |
| CritPt | 0.3% | 1.4% |
| LMArena Hard Prompts | 1372 | 1274 |
| DTBench | 77.1% | 68% |
| LMCA | 29% | 14.5% |
| Epoch Capabilities Index | 141.69 | 137.81 |
| ARC-AGI-2 | 5.9% | — |
| SimpleBench | 45.5% | — |
| ARC-AGI-1 | 40% | — |
| Chess Puzzles | — | 4% |
| EnigmaEval | 3.1% | — |
| ForecastBench | 60.2 | — |

## Math

- Claude Sonnet 4: 41.5 (#71)
- gpt-oss-20b: 37.6 (#93)

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 71.1% | 65.3% |
| Omni-MATH | 60.2% | 56.5% |
| LMArena Math | 1375 | 1317 |
| MATH Level 5 | 84.4% | — |
| FrontierMath (Feb 2025 set) | 4.1% | — |
| FrontierMath Tier 4 (v1) | 0% | — |

## Knowledge

- Claude Sonnet 4: 41.6 (#99)
- gpt-oss-20b: 34.6 (#186)

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| GPQA Diamond | 79.2% | 60.8% |
| MMLU-Pro | 84.3% | 74% |
| GPQA (HELM) | 70.6% | 59.4% |
| LMArena Expert | 1371 | 1258 |
| Humanity's Last Exam | 7.8% | — |
| Confabulations | 13.2% | — |
| Vectara Hallucination Rate | 10.3% | — |

## Multimodal

- Claude Sonnet 4: 25.6 (#116)
- gpt-oss-20b: —

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| LMArena Vision | 1191 | — |
| GeoBench | 37% | — |
| VPCT | 34% | — |
| MindCube | 44.8% | — |

## Multilingual

- Claude Sonnet 4: 46.9 (#151)
- gpt-oss-20b: 42.4 (#192)

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| LMArena Non-English | 1333 | 1269 |
| LMArena Chinese | 1351 | 1311 |
| LMArena German | 1331 | 1249 |
| LMArena Japanese | 1301 | 1245 |
| LMArena Korean | 1293 | 1239 |
| LMArena Russian | 1355 | 1279 |
| LMArena Spanish | 1356 | 1267 |
| LMArena French | 1364 | — |

## Instruction Following

- Claude Sonnet 4: 71.8 (#141)
- gpt-oss-20b: 62.0 (#235)

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| IFEval | 84% | 73.2% |
| LMArena Instruction Following | 1375 | 1236 |

## Long Context

- Claude Sonnet 4: 33.6 (#254)
- gpt-oss-20b: 37.9 (#203)

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| LMArena Longer Query | 1398 | 1251 |
| Fiction.LiveBench | 46.9% | — |

## Writing & Preference

- Claude Sonnet 4: 57.3 (#127)
- gpt-oss-20b: 35.8 (#260)

| Benchmark | Claude Sonnet 4 | gpt-oss-20b |
|---|---|---|
| LMArena Text | 1351 | 1288 |
| LMArena Creative Writing | 1345 | 1201 |
| EQ-Bench Creative Writing | 1483 | 666 |
| WildBench | 83.8% | 73.7% |
| LMArena Multi-Turn | 1376 | 1268 |
| Short-Story Creative Writing | 81.4% | — |

## FAQ

### Is Claude Sonnet 4 better than gpt-oss-20b?

Claude Sonnet 4 is the stronger model overall, scoring 40.6 to 32.1 on the Noometry Index. gpt-oss-20b costs 167× less per token, which makes it the better buy when Claude Sonnet 4's lead doesn't matter for your workload.

### Which is cheaper, Claude Sonnet 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; Claude Sonnet 4 lists at $3 and $15.

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

Claude Sonnet 4 scores higher on coding benchmarks: 44.0 versus 37.1 in the Noometry coding category.

### Which has the bigger context window?

Claude Sonnet 4 does, with 200K tokens against 131K.

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

32 benchmarks have published results for both models. Claude Sonnet 4 has 58 scored results on Noometry and gpt-oss-20b has 34.