GPT-4o vs gpt-oss-20b
gpt-oss-20b costs 66.7x less per output token than GPT-4o ($0.15 vs $10.00 per million). Updated October 21, 2026; 14 benchmarks have a result for both.
GPT-4o or gpt-oss-20b: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick GPT-4o for work measured by LiveCodeBench (Artificial Analysis run) or Terminal-Bench Hard (Artificial Analysis run), where it leads.
- Shared results
- 14
- GPT-4o only
- 13
- gpt-oss-20b only
- 48
- Like-for-like categories
- 7 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
No call
Only gpt-oss-20b has a AA Intelligence Index result, 10.0.
Directional only
Reasoning
GPQA Diamond
No call
Only gpt-oss-20b has a GPQA Diamond result, 68.9%.
Directional only
Math
AIME
No call
Only gpt-oss-20b has a AIME result, 86.0%.
Directional only
Knowledge
MMLU Pro
No call
Only gpt-oss-20b has a MMLU Pro result, 71.6%.
Directional only
Web design
Design Arena: websites
gpt-oss-20b
gpt-oss-20b leads, 852 against 831.
Same board, both measured
Long documents
Listed context window
gpt-oss-20b
gpt-oss-20b takes 131K tokens against 128K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
gpt-oss-20b
GPT-4o $0.0075, gpt-oss-20b $0.0001 per turn.
List prices
At a glance
- Coding index
- GPT-4o not benchmarked, gpt-oss-20b not benchmarked.
- Widest category
- Math, 83.3 points apart on average over 2 shared benchmarks.
- Widest single gap
- AIME 2025 (Artificial Analysis run): GPT-4o 6.0% against gpt-oss-20b 89.3%, 83.3 points apart on a 0 to 100 scale.
- Output price
- GPT-4o $10.00 against gpt-oss-20b $0.15 per million output tokens, 66.7x apart.
- Context window
- GPT-4o takes 128K tokens, gpt-oss-20b 131K.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | GPT-4o | Bars | gpt-oss-20b | Ties |
|---|---|---|---|---|
| Coding | 0 | 2 | ||
| Agentic work | 0 | 1 | ||
| Reasoning | 0 | 2 | ||
| Math | 0 | 2 | ||
| Knowledge | 0 | 0 | 1 tie | |
| Web design | 0 | 2 | ||
| General | 2 | 2 | ||
| All shared boards | 2 | 11 | 1 tie |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index02
| Benchmark | GPT-4o | gpt-oss-20b | Leader |
|---|---|---|---|
| SWE-bench Verified | 38.8% | Not measured | Only GPT-4o measured |
| SWE-bench Lite | 38.0% | Not measured | Only GPT-4o measured |
| SWE-bench (full test set) | 18.8% | Not measured | Only GPT-4o measured |
| SWE-bench Multimodal | 30.4% | Not measured | Only GPT-4o measured |
| LiveCodeBench (Vals AI) | Not measured | 80.4% | Only gpt-oss-20b measured |
| LiveCodeBench (Artificial Analysis run) | 30.9% | 77.7% | gpt-oss-20b |
| SciCode | Not measured | 38.9% | Only gpt-oss-20b measured |
| Terminal-Bench Hard (Artificial Analysis run) | 8.3% | 10.6% | gpt-oss-20b |
| Terminal-Bench | Not measured | 3.4% | Only gpt-oss-20b measured |
| WeirdML | Not measured | 40.9% | Only gpt-oss-20b measured |
| ALE-Bench | Not measured | 566.0 | Only gpt-oss-20b measured |
| LMArena Text: coding | Not measured | 1307 | Only gpt-oss-20b measured |
| LMArena Text: software and IT | Not measured | 1311 | Only gpt-oss-20b measured |
| SciCode | 1.5% | Not measured | Only GPT-4o measured |
| Neither model has a result on the other 25 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | GPT-4o | gpt-oss-20b | Leader |
|---|---|---|---|
| AA Intelligence Index | Not measured | 10.0 | Only gpt-oss-20b measured |
| Epoch Capabilities Index | Not measured | 137.7 | Only gpt-oss-20b measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | GPT-4o | gpt-oss-20b | Leader |
|---|---|---|---|
| τ²-Bench Telecom | 25.1% | 60.2% | gpt-oss-20b |
| GDPval | 9.9% | Not measured | Only GPT-4o measured |
| BrowseComp-Plus | Not measured | 50.6% | Only gpt-oss-20b measured |
| Neither model has a result on the other 38 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-4o | gpt-oss-20b | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 68.9% | Only gpt-oss-20b measured |
| GPQA Diamond (Artificial Analysis run) | 54.3% | 68.8% | gpt-oss-20b |
| Humanity's Last Exam (Artificial Analysis run) | 2.4% | 11.0% | gpt-oss-20b |
| GPQA Diamond (Epoch AI run) | Not measured | 60.8% | Only gpt-oss-20b measured |
| ARC-AGI | 4.5% | Not measured | Only GPT-4o measured |
| ARC-AGI-2 | 0.0% | Not measured | Only GPT-4o measured |
| LMArena Text: hard prompts | Not measured | 1274 | Only gpt-oss-20b measured |
| LMArena Text: expert | Not measured | 1258 | Only gpt-oss-20b measured |
| LMArena Text: hard prompts in English | Not measured | 1292 | Only gpt-oss-20b measured |
| Kagi LLM Benchmark | Not measured | 53.2% | Only gpt-oss-20b measured |
| Neither model has a result on the other 12 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-4o | gpt-oss-20b | Leader |
|---|---|---|---|
| AIME | Not measured | 86.0% | Only gpt-oss-20b measured |
| MATH 500 (Vals AI) | Not measured | 94.2% | Only gpt-oss-20b measured |
| MGSM (Vals AI) | Not measured | 89.0% | Only gpt-oss-20b measured |
| AA Math Index | 6.0 | 89.3 | gpt-oss-20b |
| AIME 2025 (Artificial Analysis run) | 6.0% | 89.3% | gpt-oss-20b |
| OTIS Mock AIME | Not measured | 65.3% | Only gpt-oss-20b measured |
| LMArena Text: math | Not measured | 1317 | Only gpt-oss-20b measured |
| LMArena Text: mathematical work | Not measured | 1305 | Only gpt-oss-20b measured |
| Neither model has a result on the other 5 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index00
| Benchmark | GPT-4o | gpt-oss-20b | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 71.6% | Only gpt-oss-20b measured |
| LegalBench (Vals AI) | Not measured | 70.8% | Only gpt-oss-20b measured |
| CaseLaw (Vals AI) | Not measured | 43.8% | Only gpt-oss-20b measured |
| CorpFin (Vals AI) | Not measured | 53.1% | Only gpt-oss-20b measured |
| TaxEval (Vals AI) | Not measured | 63.7% | Only gpt-oss-20b measured |
| MedQA (Vals AI) | Not measured | 82.9% | Only gpt-oss-20b measured |
| MMLU Pro (Artificial Analysis run) | 74.8% | 74.8% | Tie |
| MMLU | 88.1% | Not measured | Only GPT-4o measured |
| LMArena Text: science | Not measured | 1296 | Only gpt-oss-20b measured |
| LMArena Text: medicine and healthcare | Not measured | 1310 | Only gpt-oss-20b measured |
| LMArena Text: legal and government | Not measured | 1280 | Only gpt-oss-20b measured |
| LMArena Text: business and finance | Not measured | 1284 | Only gpt-oss-20b measured |
| Neither model has a result on the other 6 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-4o | gpt-oss-20b | Leader |
|---|---|---|---|
| Design Arena: websites | 831 | 852 | gpt-oss-20b |
| Design Arena: UI components | 885 | Not measured | Only GPT-4o measured |
| Design Arena: game dev | 906 | Not measured | Only GPT-4o measured |
| Design Arena: data viz | 858 | 933 | gpt-oss-20b |
| Design Arena: 3D | 869 | Not measured | Only GPT-4o measured |
| Design Arena: all code categories | 861 | Not measured | Only GPT-4o measured |
| Neither model has a result on the other 3 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index22
| Benchmark | GPT-4o | gpt-oss-20b | Leader |
|---|---|---|---|
| IFBench | 34.3% | 65.1% | gpt-oss-20b |
| Long context reasoning (AA-LCR) | 49.3% | 34.7% | GPT-4o |
| DTBench | 64.5% | 68.0% | gpt-oss-20b |
| LMCA | 16.6% | 14.5% | GPT-4o |
| LMArena Text | Not measured | 1288 | Only gpt-oss-20b measured |
| LMArena Text: instruction following | Not measured | 1236 | Only gpt-oss-20b measured |
| LMArena Text: creative writing | Not measured | 1201 | Only gpt-oss-20b measured |
| LMArena Text: English | Not measured | 1303 | Only gpt-oss-20b measured |
| LMArena Text: non-English | Not measured | 1269 | Only gpt-oss-20b measured |
| LMArena Text: Chinese | Not measured | 1315 | Only gpt-oss-20b measured |
| LMArena Text: German | Not measured | 1255 | Only gpt-oss-20b measured |
| LMArena Text: Japanese | Not measured | 1244 | Only gpt-oss-20b measured |
| LMArena Text: Korean | Not measured | 1236 | Only gpt-oss-20b measured |
| LMArena Text: Spanish | Not measured | 1267 | Only gpt-oss-20b measured |
| LMArena Text: Russian | Not measured | 1278 | Only gpt-oss-20b measured |
| LMArena Text: Polish | Not measured | 1252 | Only gpt-oss-20b measured |
| LMArena Text: multi-turn | Not measured | 1268 | Only gpt-oss-20b measured |
| LMArena Text: longer queries | Not measured | 1250 | Only gpt-oss-20b measured |
| LMArena Text: writing and language | Not measured | 1238 | Only gpt-oss-20b measured |
| LMArena Text: entertainment and media | Not measured | 1216 | Only gpt-oss-20b measured |
| Neither model has a result on the other 26 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-4o | gpt-oss-20b | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
gpt-oss-20b costs less
Repository review
50K input + 3K output tokens
gpt-oss-20b costs less
Agent loop
200K input + 10K output tokens
gpt-oss-20b costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | OpenAI GPT-4o | OpenAI gpt-oss-20b |
|---|---|---|
| Maker | OpenAI | OpenAI |
| Weights | No public weights listed | Open weights |
| Released | May 13, 2024 | August 5, 2025 |
| Context window | 128K | 131K |
| Input $/Mtok | $2.50 | $0.03 |
| Output $/Mtok | $10.00 | $0.15 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 0 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-4o better than gpt-oss-20b for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick GPT-4o for work measured by LiveCodeBench (Artificial Analysis run) or Terminal-Bench Hard (Artificial Analysis run), where it leads.
Which is cheaper, GPT-4o or gpt-oss-20b?
GPT-4o costs $2.50 per million input tokens and $10.00 per million output tokens; gpt-oss-20b costs $0.03 and $0.15. Cost per solved task is known for neither of them.
Which has the bigger context window, GPT-4o or gpt-oss-20b?
gpt-oss-20b, with 131K tokens against 128K.
How many benchmarks is the GPT-4o vs gpt-oss-20b comparison based on?
14 benchmarks have a result for both models as of October 21, 2026. GPT-4o's coding index rests on 0 independent benchmarks and gpt-oss-20b's on 0. Where only one model has a result, the page says so instead of naming a winner.