Model comparison
GPT-4o mini vs gpt-oss-20b
gpt-oss-20b is the stronger model overall, scoring 32.5 to 25.5 on the Noometry Index.
Last verified . 30 shared benchmarks.
Summary
- They share 30 benchmarks with published results for both. GPT-4o mini scores higher in 4 categories and gpt-oss-20b in 5 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where gpt-oss-20b leads 39.4 to 10.4.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 6.9% for GPT-4o mini and 65.3% for gpt-oss-20b.
- gpt-oss-20b is cheaper at $0.018 / $0.09 per million input/output tokens, against $0.15 / $0.60 for GPT-4o mini.
- gpt-oss-20b accepts more context: 131K tokens versus 128K.
- gpt-oss-20b has downloadable open weights; the other is API-only.
Side by side
| GPT-4o mini | gpt-oss-20b | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 25.5 | 32.5 |
| Released | 2024-07-18 | 2025-08-05 |
| Weights | Proprietary | Open |
| Context window | 128K | 131K |
| Max output | 16K | 16K |
| Input $ / M tokens | $0.15 | $0.018 |
| Output $ / M tokens | $0.60 | $0.09 |
| Results tracked | 60 | 34 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding gpt-oss-20b leads
GPT-4o mini: 22.0 (#335), gpt-oss-20b: 37.6 (#192)
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| WeirdML | 11.8% | 40.9% |
| LMArena Coding | 1290 | 1306 |
| Aider Polyglot | 3.6% | — |
| SciCode | — | 34.4% |
| BigCodeBench Instruct | 46.1% | — |
| LiveBench Coding | 43.1% | — |
| BigCodeBench Complete | 57.4% | — |
| ALE-Bench | — | 566.05 |
| HumanEval+ | 83.5% | — |
| MBPP+ | 72.2% | — |
Agentic & Tool Use GPT-4o mini leads
GPT-4o mini: 27.5 (#101), gpt-oss-20b: 9.3 (#154)
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| Terminal-Bench | — | 3.4% |
| BALROG | 17.4% | — |
Reasoning gpt-oss-20b leads
GPT-4o mini: 8.7 (#347), gpt-oss-20b: 19.3 (#261)
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| Kagi LLM Benchmark | 28.8% | 53.2% |
| Chess Puzzles | 0% | 4% |
| LMArena Hard Prompts | 1267 | 1274 |
| DTBench | 54.4% | 68% |
| LMCA | 10.4% | 14.5% |
| Epoch Capabilities Index | 126.56 | 137.82 |
| ARC-AGI-2 | 0% | — |
| SimpleBench | 10.7% | — |
| CritPt | — | 1.4% |
| LiveBench Reasoning | 32.8% | — |
| Mystery Game Puzzles | 12% | — |
| LiveBench Data Analysis | 50% | — |
| LiveBench | 41.3% | — |
| PIQA | 88.7% | — |
Math gpt-oss-20b leads
GPT-4o mini: 10.4 (#314), gpt-oss-20b: 39.4 (#103)
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 6.9% | 65.3% |
| Omni-MATH | 28% | 56.5% |
| LMArena Math | 1267 | 1317 |
| FrontierMath (Tiers 1-3) | 0.7% | — |
| LiveBench Math | 36.3% | — |
| MATH Level 5 | 52.6% | — |
| GSM8K | 91.3% | — |
Knowledge gpt-oss-20b leads
GPT-4o mini: 17.7 (#284), gpt-oss-20b: 34.6 (#195)
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| GPQA Diamond | 37.7% | 60.8% |
| MMLU-Pro | 60.3% | 74% |
| GPQA (HELM) | 36.8% | 59.4% |
| LMArena Expert | 1235 | 1258 |
| SimpleQA Verified | 8.3% | — |
| Confabulations | 37.2% | — |
| BoolQ | 88.7% | — |
| MMLU | 81.8% | — |
Multimodal Not comparable
GPT-4o mini: 25.9 (#122), gpt-oss-20b: —
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| LMArena Vision | 1066 | — |
| Video-MME | 64.8% | — |
| GeoBench | 64% | — |
| VPCT | 34% | — |
Multilingual Too close to call
GPT-4o mini: 42.0 (#199), gpt-oss-20b: 42.2 (#197)
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| LMArena Non-English | 1266 | 1268 |
| LMArena Chinese | 1265 | 1314 |
| LMArena German | 1272 | 1255 |
| LMArena Japanese | 1216 | 1244 |
| LMArena Korean | 1195 | 1236 |
| LMArena Russian | 1275 | 1278 |
| LMArena Spanish | 1276 | 1267 |
| LMArena French | 1297 | — |
Instruction Following Too close to call
GPT-4o mini: 61.9 (#239), gpt-oss-20b: 61.8 (#240)
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| IFEval | 78.2% | 73.2% |
| LMArena Instruction Following | 1258 | 1236 |
| LiveBench Instruction Following | 56.8% | — |
Long Context GPT-4o mini leads
GPT-4o mini: 39.1 (#186), gpt-oss-20b: 37.9 (#209)
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| LMArena Longer Query | 1289 | 1250 |
Writing & Preference GPT-4o mini leads
GPT-4o mini: 39.5 (#248), gpt-oss-20b: 35.5 (#265)
| Benchmark | GPT-4o mini | gpt-oss-20b |
|---|---|---|
| LMArena Text | 1286 | 1287 |
| LMArena Creative Writing | 1268 | 1201 |
| EQ-Bench Creative Writing | 873 | 666 |
| WildBench | 79.1% | 73.7% |
| LMArena Multi-Turn | 1285 | 1268 |
| Short-Story Creative Writing | 67.2% | — |
| LiveBench Language | 28.6% | — |
Frequently asked questions
Is GPT-4o mini better than gpt-oss-20b?
gpt-oss-20b is the stronger model overall, scoring 32.5 to 25.5 on the Noometry Index.
Which is cheaper, GPT-4o mini 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-4o mini lists at $0.15 and $0.60.
Is GPT-4o mini or gpt-oss-20b better for coding?
gpt-oss-20b scores higher on coding benchmarks: 37.6 versus 22.0 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-4o mini and gpt-oss-20b share?
30 benchmarks have published results for both models. GPT-4o mini has 60 scored results on Noometry and gpt-oss-20b has 34.