Model comparison
GPT-5.4 nano vs gpt-oss-20b
GPT-5.4 nano is the stronger model overall, scoring 41.9 to 32.1 on the Noometry Index. gpt-oss-20b costs 13× less per token, which makes it the better buy when GPT-5.4 nano's lead doesn't matter for your workload.
Last verified . 27 shared benchmarks.
Summary
- They share 27 benchmarks with published results for both. GPT-5.4 nano scores higher in 8 categories and gpt-oss-20b in 0 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where GPT-5.4 nano leads 55.9 to 35.8.
- The biggest single-benchmark swing is Chess Puzzles: 30% for GPT-5.4 nano and 4% for gpt-oss-20b.
- gpt-oss-20b is cheaper at $0.018 / $0.09 per million input/output tokens, against $0.20 / $1.25 for GPT-5.4 nano.
- GPT-5.4 nano accepts more context: 400K tokens versus 131K.
- gpt-oss-20b has downloadable open weights; the other is API-only.
Side by side
| GPT-5.4 nano | gpt-oss-20b | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 41.9 | 32.1 |
| Released | 2026-03-17 | 2025-08-05 |
| Weights | Proprietary | Open |
| Context window | 400K | 131K |
| Max output | 128K | 16K |
| Input $ / M tokens | $0.20 | $0.018 |
| Output $ / M tokens | $1.25 | $0.09 |
| Results tracked | 40 | 34 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.4 nano leads
GPT-5.4 nano: 43.1 (#82), gpt-oss-20b: 37.1 (#185)
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| SciCode | 46.9% | 34.4% |
| WeirdML | 49.2% | 40.9% |
| LMArena Coding | 1404 | 1307 |
| ALE-Bench | 1,005 | 566.05 |
Agentic & Tool Use Not comparable
GPT-5.4 nano: —, gpt-oss-20b: 9.2 (#146)
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| Terminal-Bench | — | 3.4% |
Reasoning GPT-5.4 nano leads
GPT-5.4 nano: 23.8 (#159), gpt-oss-20b: 18.8 (#251)
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| Kagi LLM Benchmark | 39.7% | 53.2% |
| CritPt | 9.3% | 1.4% |
| Chess Puzzles | 30% | 4% |
| LMArena Hard Prompts | 1380 | 1274 |
| DTBench | 80.3% | 68% |
| LMCA | 36.9% | 14.5% |
| Epoch Capabilities Index | 145.78 | 137.81 |
| ARC-AGI-2 | 5.7% | — |
| ARC-AGI-1 | 51.5% | — |
| Mystery Game Puzzles | 9% | — |
| ForecastBench | 57.3 | — |
Math GPT-5.4 nano leads
GPT-5.4 nano: 40.7 (#76), gpt-oss-20b: 37.6 (#93)
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 87.8% | 65.3% |
| LMArena Math | 1406 | 1317 |
| FrontierMath (Tiers 1-3) | 44.9% | — |
| FrontierMath Tier 4 | 12.2% | — |
| ProofBench | 5% | — |
| Omni-MATH | — | 56.5% |
| FrontierMath (Feb 2025 set) | 25.9% | — |
| FrontierMath Tier 4 (v1) | 6.3% | — |
Knowledge GPT-5.4 nano leads
GPT-5.4 nano: 41.8 (#94), gpt-oss-20b: 34.6 (#186)
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| GPQA Diamond | 78.5% | 60.8% |
| LMArena Expert | 1397 | 1258 |
| SimpleQA Verified | 11.7% | — |
| MMLU-Pro | — | 74% |
| Vectara Hallucination Rate | 3.1% | — |
| GPQA (HELM) | — | 59.4% |
Multimodal Not comparable
GPT-5.4 nano: 36.0 (#73), gpt-oss-20b: —
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| LMArena Vision | 1196 | — |
Multilingual GPT-5.4 nano leads
GPT-5.4 nano: 48.8 (#136), gpt-oss-20b: 42.4 (#192)
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| LMArena Non-English | 1359 | 1269 |
| LMArena Chinese | 1391 | 1311 |
| LMArena German | 1368 | 1249 |
| LMArena Japanese | 1342 | 1245 |
| LMArena Korean | 1321 | 1239 |
| LMArena Russian | 1363 | 1279 |
| LMArena Spanish | 1368 | 1267 |
| LMArena French | 1396 | — |
Instruction Following GPT-5.4 nano leads
GPT-5.4 nano: 72.1 (#139), gpt-oss-20b: 62.0 (#235)
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| LMArena Instruction Following | 1362 | 1236 |
| IFEval | — | 73.2% |
Long Context GPT-5.4 nano leads
GPT-5.4 nano: 41.6 (#131), gpt-oss-20b: 37.9 (#203)
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| LMArena Longer Query | 1365 | 1251 |
Writing & Preference GPT-5.4 nano leads
GPT-5.4 nano: 55.9 (#137), gpt-oss-20b: 35.8 (#260)
| Benchmark | GPT-5.4 nano | gpt-oss-20b |
|---|---|---|
| LMArena Text | 1372 | 1288 |
| LMArena Creative Writing | 1313 | 1201 |
| LMArena Multi-Turn | 1382 | 1268 |
| EQ-Bench Creative Writing | — | 666 |
| WildBench | — | 73.7% |
Frequently asked questions
Is GPT-5.4 nano better than gpt-oss-20b?
GPT-5.4 nano is the stronger model overall, scoring 41.9 to 32.1 on the Noometry Index. gpt-oss-20b costs 13× less per token, which makes it the better buy when GPT-5.4 nano's lead doesn't matter for your workload.
Which is cheaper, GPT-5.4 nano 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-5.4 nano lists at $0.20 and $1.25.
Is GPT-5.4 nano or gpt-oss-20b better for coding?
GPT-5.4 nano scores higher on coding benchmarks: 43.1 versus 37.1 in the Noometry coding category.
Which has the bigger context window?
GPT-5.4 nano does, with 400K tokens against 131K.
How many benchmarks do GPT-5.4 nano and gpt-oss-20b share?
27 benchmarks have published results for both models. GPT-5.4 nano has 40 scored results on Noometry and gpt-oss-20b has 34.