Inference
EN
OpenAI

GPT-OSS 20B

GPT-OSS 20B is a standard model from OpenAI in the GPT-OSS family. 20 benchmarks count toward its index, in 7 categories.

  • Open weights
  • Standard
  • Input: text
  • Released 2025-08
RankedEnough results for an index.
IndexAll categories, out of 100.41.7±4.6Median 51.0
CoverageIndex weight that has results.90%
SpeedOutput tokens per second.310/sMedian 67/s
First tokenSeconds to the first output token.0.4 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.03Median $0.60
Output / 1MUS dollars per 1M output tokens.$0.13Median $2.55
ContextMaximum tokens in one request.131KMedian 262K
EloLMArena rating and rank.1288±6 #264Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
openai/gpt-oss-20b
Max outputMaximum output tokens per request.
118K
Knowledge cutoffLast date of training data.
30 Jun 2024
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.03
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
14 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
CoreWeaveHost$0.03$0.13131K
OpenRouterRouter$0.03$0.13131K
Deep InfraHost$0.03$0.14131K
SiliconFlowHost$0.04$0.18131K
OVHcloudCloud$0.05$0.18131K
IO.NETHost$0.055$0.19464K
STACKITCloud$0.18$0.29131K
Amazon BedrockCloud$0.07$0.30131K
DatabricksCloud$0.07$0.30131K
Oracle CloudCloud$0.07$0.30128K
GroqHost$0.075$0.30131K
Cloudflare Workers AICloud$0.20$0.30128K
DigitalOceanCloud$0.05$0.45128K
RegoloHost$0.112$0.471128K

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Run locally

20.91B parameters
16-bit
44.9 GB
8-bit
24.3 GB
4-bit
14.3 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy fits in 16 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.3 counted39.8179 / 218
CodingCode writing and repair.4 counted48.9145 / 209
ReasoningLogic problems and puzzles.2 counted35.3189 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.4 counted36.0218 / 268
MultilingualTasks in many languages.1 counted37.0—
InstructionTasks with strict prompt rules.1 counted49.9—
MathMath problems.5 counted45.9105 / 168

Results

20 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
American Invitational Mathematics Examination 2024Math92.1%4 / 27
MATH 500Math94.2% ±1.010 / 59
EuroEval SwedishMultilingual55.5% ±0.250 / 195
American Invitational Mathematics Examination 2025Math91.7%15 / 53
EuroEval GermanMultilingual45.8% ±0.457 / 191
EuroEval PolishMultilingual50.2% ±0.366 / 209
EuroEval SwedishMultilingual52.0% ±0.463 / 195
EuroEval ItalianMultilingual51.0% ±0.358 / 174
EuroEval SpanishMultilingual45.8% ±0.359 / 176
EuroEval PolishMultilingual48.6% ±0.372 / 209
AIMEMath86.0% ±1.133 / 95
EuroEval PortugueseMultilingual50.6% ±0.466 / 182
EuroEval SpanishMultilingual44.2% ±0.464 / 176
EuroEval PortugueseMultilingual49.7% ±0.367 / 182
EuroEval GermanMultilingual42.5% ±0.671 / 191
Artificial Analysis IFBenchInstruction65.1%55 / 146
EuroEval SwedishMultilingual48.4% ±0.474 / 195
EuroEval DutchMultilingual49.9% ±0.370 / 175
EuroEval ItalianMultilingual47.8% ±0.372 / 174
EuroEval FrenchMultilingual52.9% ±0.376 / 178
EuroEval FrenchMultilingual51.7% ±0.381 / 178
EuroEval PolishMultilingual43.5% ±0.397 / 209
EuroEval GermanMultilingual38.3% ±0.689 / 191
LiveCodeBenchCoding80.4% ±1.169 / 142
EuroEval ItalianMultilingual43.3% ±0.588 / 174
Critical Physics TasksReasoning1.4%113 / 214
Massive Multitask Language UnderstandingKnowledge85.3%18 / 34
EuroEval FrenchMultilingual44.8% ±0.699 / 178
EuroEval DutchMultilingual40.5% ±0.698 / 175
τ²-Bench Tool-Agent-User EvaluationAgentic60.2%86 / 149
EuroEval SpanishMultilingual32.9% ±0.6102 / 176
EuroEval PortugueseMultilingual40.7% ±0.4106 / 182
OTIS Mock AIME 2024-2025Math50.8% ±6.7122 / 200
Graduate-Level Google-Proof Q&AKnowledge71.5%87 / 141
Artificial Analysis Humanity's Last ExamKnowledge11.0%141 / 222
Artificial Analysis GPQA DiamondKnowledge68.8%134 / 199
GPQA Diamond · Vals AIKnowledge68.9% ±2.597 / 135
GPQA diamond · Epoch AIKnowledge46.0% ±3.6161 / 217
Software Engineering Benchmark VerifiedCoding60.7%86 / 114
GDPval-AA normalizedAgentic0.0%101 / 133
Artificial Analysis Long Context ReasoningReasoning34.7%175 / 217
Artificial Analysis SciCodeCoding38.9%92 / 114
MMLU ProKnowledge71.6% ±0.4115 / 135
Humanity's Last ExamKnowledge10.9%72 / 77
React Native EvalsCoding71.0%16 / 16
APEX-Agents-AAAgentic0.7%30 / 30
MGSMMultilingual89.0% ±1.950 / 74
MedQAHealthcare82.9% ±0.365 / 95
CorpFin v2Finance53.1% ±1.099 / 131
LegalBenchLegal70.8% ±0.6122 / 149
TaxEval v2Finance63.7% ±0.9119 / 143
CaseLaw v2Legal43.8% ±0.953 / 53

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedEuroEvalMITEpoch AICC BYmodels.devMIT

History

5 releases
Index42.1 → 41.7
23 Oct 202612 Nov 2026
Output / 1M$0.09 → $0.13
23 Oct 202612 Nov 2026

Questions

How much does GPT-OSS 20B cost?
$0.03 per 1M input tokens, $0.13 per 1M output tokens.
How good is GPT-OSS 20B?
Its index is 41.7 out of 100. That is place 201 of 283 ranked models.
What is GPT-OSS 20B best at?
Math: 45.9 out of 100, place 105 of 168.
How fast is GPT-OSS 20B?
310 output tokens per second, 0.4 s to the first token.
What is the context window of GPT-OSS 20B?
131K tokens.
Is GPT-OSS 20B open source?
Yes. Its weights are published.
When was GPT-OSS 20B released?
August 2025.

Badge

GPT-OSS 20B on Inference360: place 201 of 283, index 41.7
HTML
<a href="https://inference360.ai/models/gpt-oss-20b"><img src="https://inference360.ai/badges/gpt-oss-20b.svg" alt="GPT-OSS 20B on Inference360: place 201 of 283, index 41.7" height="20"></a>
Markdown
[![GPT-OSS 20B on Inference360: place 201 of 283, index 41.7](https://inference360.ai/badges/gpt-oss-20b.svg)](https://inference360.ai/models/gpt-oss-20b)

Compare with

GLM-4.641.8Granite 4.2 30B41.6GPT-5 nano41.9LFM3-1.7B41.5GPT-OSS 120B43.4

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeNemotron 3 Ultra50.8 · Free

More from OpenAI

All models
GPT-OSS 120B43.4GPT-5 (high)53.6GPT-5 (medium)53.0GPT-5 mini48.4GPT-5 nano41.9