GPT OSS 20B
An open-weight 21B parameter model released under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for lower-latency inference and deployability on consumer or single-GPU hardware. The model is trained in OpenAI's Harmony response format and supports reasoning level configuration, fine-tuning, and agentic capabilities including function calling, tool use, and structured outputs.
- Reasoning
Added Aug 5, 2025
Model weightsPricing
Auto routing · per 1M tokens- Input
- $0.018
- Output
- $0.09
- Cache read
- $0.009
Specifications
- Context window
- 131.1K
- Max output
- 16.4K
- Parameters
- 20B / 3.6B
- Total / active
Benchmarks
Benchmarks
Sourced from LMArena.
Arena Score
1317.5
Overall Rank
#272 / 421
Votes
10,396
Confidence Interval
1311.1 - 1323.9
Category Scores
Coding
#253 / 416
2,111 votes
1368.3
Math
#228 / 404
677 votes
1335.0
Longer Query
#288 / 399
2,100 votes
1303.2
Creative Writing
#309 / 419
1,229 votes
1241.0
Instruction Following
#291 / 421
2,534 votes
1281.3
Hard Prompts
#271 / 421
4,706 votes
1322.9
Additional Categories21
Japanese
#201 / 288
246 votes
1255.4
Korean
#203 / 289
227 votes
1275.8
Polish
#231 / 243
807 votes
1275.5
Industry Mathematical
#235 / 402
583 votes
1329.9
Industry Medicine And Healthcare
#236 / 391
562 votes
1350.8
Chinese
#240 / 395
665 votes
1351.7
Spanish
#244 / 304
283 votes
1303.7
German
#246 / 320
217 votes
1273.8
Industry Business And Management And Financial Operations
#248 / 414
1,832 votes
1321.6
Industry Software And It Services
#250 / 421
3,586 votes
1357.5
Expert
#252 / 371
504 votes
1315.9
Russian
#253 / 385
751 votes
1309.6
Non English
#265 / 421
5,886 votes
1298.9
Hard Prompts English
#266 / 419
2,074 votes
1337.3
Exclude Ties
#268 / 421
7,607 votes
1262.0
English
#270 / 421
4,499 votes
1331.3
Industry Life And Physical And Social Science
#274 / 419
1,664 votes
1328.4
Industry Legal And Government
#283 / 393
678 votes
1321.6
Multi Turn
#286 / 419
1,775 votes
1289.2
Industry Writing And Literature And Language
#294 / 420
2,164 votes
1275.4
Industry Entertainment And Sports And Media
#295 / 419
1,726 votes
1256.2
Published 2026-11-11 · Matched as gpt-oss-20b · CC BY 4.0
LMArena DatasetLeaderboard results
| Benchmark | Score | Rank on NanoGPT |
|---|---|---|
Epoch AI Benchmarking Hub Epoch AI data, CC BY 4.0 · Updated Nov 14, 2026 | ||
Epoch Capabilities Index | 135.0 | #101 of 115 |
GPQA Diamond | 60.8% | #100 of 119 |
OTIS Mock AIME 2024–2025 | 65.3% | #95 of 121 |
Creative Writing v3 | 665 | #95 of 96 |
Longform Writing | 21.2 | #97 of 97 |
EQ-Bench 3 | 985 | #42 of 49 |
UGI Leaderboard UGI Leaderboard data, Apache-2.0 · Updated Oct 16, 2026 | ||
UGI Score | 12.4 | #144 of 149 |
UGI Willingness | 1.8 | #133 of 149 |
UGI Writing | 24.6 | #139 of 149 |
Filtered and ranked to models on NanoGPT.
Providers
Choose or restrict providers with the API. Provider-selection prices include the selection markup. Provider selection guide · Pricing API (JSON, USD per 1K tokens)
 | $0.018 | $0.09 | $0.009 | N/A | N/A |
| $0.018 | $0.09 | $0.009 | N/A | N/A |
| $0.00945 | $0.0473 | $0.0047 | N/A | N/A | |
| $0.0189 | $0.0945 | $0.00945 | N/A | N/A | |
| $0.021$0.0189 | $0.105$0.0945 | N/A | N/A | N/A | |
| $0.021 | $0.105 | N/A | N/A | N/A | |
| $0.0231 | $0.0346 | $0.021 | N/A | N/A | |
| $0.0305 | $0.147 | $0.0305 | N/A | N/A | |
| $0.0315 | $0.136 | $0.0315 | N/A | N/A | |
| $0.0315 | $0.147 | N/A | N/A | N/A | |
| $0.0315 | $0.157 | $0.021 | N/A | N/A | |
| $0.032 | $0.149 | N/A | N/A | N/A | |
| $0.042 | $0.189 | N/A | N/A | N/A | |
| $0.0735 | $0.315 | $0.0367 | N/A | N/A | |
| $0.0787 | $0.315 | $0.0394 | N/A | N/A |
Batch rows for this model use normal routing and cost the same as standard API requests. Set provider in a row to choose a provider.