Rankings / OpenAI
GPT OSS 20B
open weightsreleased 2025-08-05
BenchAtlas Index
as of 2026-09-11
high effort
rank #333
10 families · 5 categories · high
base configuration
rank #367
8 families · 4 categories · medium
low effort
rank #393
10 families · 5 categories · high
Benchmark evidence
69 results
agentic coding
| τ²-Bench | high effort | 60.2 % | independent | Artificial Analysis |
| τ²-Bench | low effort | 50.3 % | independent | Artificial Analysis |
| τ²-Bench subset=banking | high effort | 7.0 % | independent | Artificial Analysis |
coding
| LiveCodeBench v6 implementation=artificial-analysis | high effort | 77.7 % | independent | Artificial Analysis |
| LiveCodeBench v6 implementation=artificial-analysis | low effort | 65.2 % | independent | Artificial Analysis |
| SciCode | high effort | 38.9 % | independent | Artificial Analysis |
| SciCode | low effort | 34.0 % | independent | Artificial Analysis |
external indices
| AA Coding Index | high effort | 20.7 points | independent | Artificial Analysis |
| Intelligence Index v4.1 | low effort | 10.0 points | independent | Artificial Analysis |
| Intelligence Index v4.1 | high effort | 9.0 points | independent | Artificial Analysis |
| AA Math Index | high effort | 89.3 points | independent | Artificial Analysis |
| AA Math Index | low effort | 62.3 points | independent | Artificial Analysis |
| ECI | low effort | 137.8 points | independent | Epoch AI Benchmarking Hub |
| ECI | 137.8 points | independent | Epoch AI Benchmarking Hub | |
| ECI | medium effort | 137.8 points | independent | Epoch AI Benchmarking Hub |
| ECI | high effort | 137.8 points | independent | Epoch AI Benchmarking Hub |
factuality
| MASK contamination=Potential contamination warning: This model was evaluated after the public release of MASK, allowing model builder access to the prompts and solutions. | 86.5 % 84.4–88.5 | independent | Scale Labs |
human preference
| Coding (style control)older version arena=text · category=coding · style_control=true | 1369.2 1356.1–1382.2 | community | LMArena Leaderboard Dataset | |
| Industry Software And It Services (style control)older version arena=text · category=industry_software_and_it_services · style_control=true | 1358.1 1348.0–1368.1 | community | LMArena Leaderboard Dataset | |
| Chinese (style control)older version arena=text · category=chinese · style_control=true | 1351.9 1328.9–1374.9 | community | LMArena Leaderboard Dataset | |
| Industry Medicine And Healthcare (style control)older version arena=text · category=industry_medicine_and_healthcare · style_control=true | 1350.0 1324.2–1375.7 | community | LMArena Leaderboard Dataset | |
| Hard Prompts English (style control)older version arena=text · category=hard_prompts_english · style_control=true | 1336.6 1323.2–1349.9 | community | LMArena Leaderboard Dataset | |
| Math (style control)older version arena=text · category=math · style_control=true | 1335.3 1313.2–1357.5 | community | LMArena Leaderboard Dataset | |
| English (style control)older version arena=text · category=english · style_control=true | 1331.0 1321.9–1340.1 | community | LMArena Leaderboard Dataset | |
| Industry Mathematical (style control)older version arena=text · category=industry_mathematical · style_control=true | 1329.0 1304.9–1353.2 | community | LMArena Leaderboard Dataset | |
| Industry Life And Physical And Social Science (style control)older version arena=text · category=industry_life_and_physical_and_social_science · style_control=true | 1326.7 1311.6–1341.9 | community | LMArena Leaderboard Dataset | |
| Industry Business And Management And Financial Operations (style control)older version arena=text · category=industry_business_and_management_and_financial_operations · style_control=true | 1322.7 1308.8–1336.7 | community | LMArena Leaderboard Dataset | |
| Hard Prompts (style control)older version arena=text · category=hard_prompts · style_control=true | 1322.2 1313.3–1331.1 | community | LMArena Leaderboard Dataset | |
| Industry Legal And Government (style control)older version arena=text · category=industry_legal_and_government · style_control=true | 1320.4 1296.8–1344.0 | community | LMArena Leaderboard Dataset | |
| Overall (style control) arena=text · category=overall · style_control=true | 1317.1 1310.7–1323.5 | community | LMArena Leaderboard Dataset | |
| Expert (style control)older version arena=text · category=expert · style_control=true | 1313.6 1286.0–1341.3 | community | LMArena Leaderboard Dataset | |
| Russian (style control)older version arena=text · category=russian · style_control=true | 1308.2 1287.1–1329.3 | community | LMArena Leaderboard Dataset | |
| Longer Query (style control)older version arena=text · category=longer_query · style_control=true | 1301.8 1288.5–1315.2 | community | LMArena Leaderboard Dataset | |
| Non English (style control)older version arena=text · category=non_english · style_control=true | 1298.6 1290.5–1306.6 | community | LMArena Leaderboard Dataset | |
| Multi Turn (style control)older version arena=text · category=multi_turn · style_control=true | 1289.3 1274.9–1303.8 | community | LMArena Leaderboard Dataset | |
| Instruction Following (style control)older version arena=text · category=instruction_following · style_control=true | 1281.2 1269.1–1293.4 | community | LMArena Leaderboard Dataset | |
| Polish (style control)older version arena=text · category=polish · style_control=true | 1274.8 1252.5–1297.0 | community | LMArena Leaderboard Dataset | |
| Industry Writing And Literature And Language (style control)older version arena=text · category=industry_writing_and_literature_and_language · style_control=true | 1274.0 1260.8–1287.2 | community | LMArena Leaderboard Dataset | |
| Exclude Ties (style control)older version arena=text · category=exclude_ties · style_control=true | 1261.6 1252.5–1270.7 | community | LMArena Leaderboard Dataset | |
| Industry Entertainment And Sports And Media (style control)older version arena=text · category=industry_entertainment_and_sports_and_media · style_control=true | 1254.7 1239.6–1269.8 | community | LMArena Leaderboard Dataset | |
| Creative Writing (style control)older version arena=text · category=creative_writing · style_control=true | 1239.6 1221.3–1257.9 | community | LMArena Leaderboard Dataset |
knowledge science
| GPQA Diamond implementation=vals-ai | 68.9 % | independent | Vals AI | |
| GPQA Diamond implementation=artificial-analysis | high effort | 68.8 % | independent | Artificial Analysis |
| GPQA Diamond implementation=artificial-analysis | low effort | 61.1 % | independent | Artificial Analysis |
| GPQA Diamond | medium effort | 60.8 % | independent | Epoch AI Benchmarking Hub 2026-08-27 |
| GPQA Diamond | low effort | 53.2 % | independent | Epoch AI Benchmarking Hub 2026-08-27 |
| GPQA Diamond | high effort | 46.0 % | independent | Epoch AI Benchmarking Hub 2026-08-06 |
| Humanity's Last Exam implementation=artificial-analysis | high effort | 11.0 % | independent | Artificial Analysis |
| Humanity's Last Exam implementation=artificial-analysis | low effort | 5.3 % | independent | Artificial Analysis |
| MMLU-Pro implementation=artificial-analysis | high effort | 74.8 % | independent | Artificial Analysis |
| MMLU-Pro implementation=artificial-analysis | low effort | 71.8 % | independent | Artificial Analysis |
| MMLU-Pro implementation=vals-ai | 71.6 % | independent | Vals AI |
long context instruction
| AA-LCR | high effort | 34.7 % | independent | Artificial Analysis |
| AA-LCR | low effort | 30.7 % | independent | Artificial Analysis |
| IFBench | high effort | 65.1 % | independent | Artificial Analysis |
| IFBench | low effort | 57.8 % | independent | Artificial Analysis |
professional
| CaseLaw | 43.8 % | independent | Vals AI | |
| CorpFin | 53.1 % | independent | Vals AI | |
| LegalBench | 70.8 % | independent | Vals AI | |
| MedQA | 82.9 % | independent | Vals AI | |
| TaxEval | 63.7 % | independent | Vals AI |
reasoning math
| AIME year=2025 · implementation=artificial-analysis | high effort | 89.3 % | independent | Artificial Analysis |
| AIME implementation=vals-ai | 86.0 % | independent | Vals AI | |
| AIME year=2025 · implementation=artificial-analysis | low effort | 62.3 % | independent | Artificial Analysis |
| MATH-500 implementation=vals-ai | 94.2 % | independent | Vals AI | |
| OTIS Mock AIME 2024–2025 | medium effort | 65.3 % | independent | Epoch AI Benchmarking Hub 2026-08-27 |
| OTIS Mock AIME 2024–2025 | high effort | 53.9 % | independent | Epoch AI Benchmarking Hub 2026-08-07 |
| OTIS Mock AIME 2024–2025 | high effort | 50.8 % | independent | Epoch AI Benchmarking Hub 2026-08-27 |
| OTIS Mock AIME 2024–2025 | low effort | 40.3 % | independent | Epoch AI Benchmarking Hub 2026-08-27 |
Agent + model results
systems, not bare-model scores
| agent + model Artificial Analysis harness + GPT OSS 20B | Terminal-Bench 2.1 | 13.9 % | independent | Artificial Analysis |
| agent + model Artificial Analysis harness + GPT OSS 20B | Terminal-Bench Hard | 10.6 % | independent | Artificial Analysis |
| agent + model Artificial Analysis harness + GPT OSS 20B | Terminal-Bench Hard | 4.5 % | independent | Artificial Analysis |
These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.