NEWHow AI models draw the world map · Gecko Tests
Beta
Home/Models/gpt-oss-20b
OpenAI logo

gpt-oss-20b

by OpenAI · Released Aug 2025

Open Source
45.0
avg score
Rank #171
Compare
Better than 45% of all models
Context
131K tokens (~66 books)
Input $/1M
$0.02
Output $/1M
$0.09
Type
text
License
Open Source
Benchmarks
29 tested
Data as of
About

gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for...

Tested on 29 benchmarks · BenchGecko score 45.0. Top scores: Chatbot Arena Elo — Overall (1317.6%), OpenCompass — IFEval (88.9%), OpenCompass — AIME2025 (87.9%).

Capabilities
coding
37.6
#145 globally
reasoning
73.7
#33 globally
math
69.9
#46 globally
knowledge
47.8
#146 globally
language
81.1
#55 globally
speed
30.0
#85 globally
general
31.9
#118 globally
Benchmark Scores
Compare All
Tested on 29 benchmarks · Ranked across 8 categories
Score Distribution (all 312 models)
0255075100
▲ You are here
codingCompare coding →
OpenCompass — LiveCodeBenchV6

OpenCompass Live Code Bench v6. Fresh competitive programming problems to evaluate code generation without memorization.

68.4·
WeirdML

Unusual and adversarial machine learning challenges. Tests robustness of reasoning about edge cases in ML systems.

40.9·
Terminal Bench

Complex terminal-based engineering tasks. Models must use command-line tools, navigate filesystems, and debug systems through shell interaction.

3.4·
reasoningCompare reasoning →
HELM — WildBench

Stanford HELM WildBench evaluation. Tests reasoning on challenging real-world tasks.

73.7·
mathCompare math →
OpenCompass — AIME2025

OpenCompass evaluation on AIME 2025 problems. Tests mathematical reasoning on fresh competition problems.

87.9·
OTIS Mock AIME 2024-2025

Mock AIME (American Invitational Mathematics Exam) problems from OTIS. Tests mathematical competition performance.

65.2·
HELM — Omni-MATH

Stanford HELM evaluation of mathematical reasoning across diverse problem types.

56.5·
Quick compare:Alibaba Qwen logovs Qwen3 235B A22B Instruct 2507Google DeepMind logovs Gemini 2.0 Flash LiteAlibaba Qwen logovs Qwen3.7 Flash
Excellent (85+) Good (70-85) Average (50-70) Below (<50)
Recently Happened
gpt-oss-20b pricing dropped 40%
Sep 23, 2026
gpt-oss-20b pricing increased 27%
Apr 11, 2026
Similar Models
Alibaba Qwen logo
Qwen3 235B A22B Instruct 2507
Alibaba Qwen
44.9$0.09/1M
Google DeepMind logo
Gemini 2.0 Flash Lite
Google DeepMind
45.3$0.07/1M
Alibaba Qwen logo
Qwen3.7 Flash
Alibaba Qwen
44.6$0.03/1M
Links
Info
OpenAIPricing explorerDevelopers · API
Research
Technical Report
Documentation
API DocsPlayground
Community
@OpenAI
Source Code
GitHubHugging Face
BenchGecko API
gpt-oss-20b
Specifications
  • Typetext
  • Context131K tokens (~66 books)
  • ReleasedAug 2025
  • LicenseOpen Source
  • StatusActive
  • Cost / Message~$0.000
Available On
OpenAI logoOpenAI$0.02
Categories
codingreasoningmathknowledgelanguagespeedgeneral
Learn More
context-windowtransformeropen weightstokens
Share & Export
Tweet
Related Models
Alibaba Qwen logoQwen3 235B A22B Instruct 2507
44.9
Google DeepMind logoGemini 2.0 Flash Lite
45.3
Alibaba Qwen logoQwen3.7 Flash
44.6
Anthropic logoClaude Sonnet 4.5
44.6
gpt-oss-20b is an open-source text AI model by OpenAI, released in August 2025. It has an average benchmark score of 45.0. Context window: 131K tokens.

Related Models

Qwen3 235B A22B Instruct 2507 · Alibaba QwenGemini 2.0 Flash Lite · Google DeepMindQwen3.7 Flash · Alibaba QwenClaude Sonnet 4.5 · AnthropicGrok 3 · xAI

Benchmarks

Chatbot Arena Elo — OverallOpenCompass — IFEvalOpenCompass — AIME2025HELM — MMLU-ProHELM — WildBench

Related Pages

OpenAI · ProviderOpenAI · EconomyAll ModelsCompare ModelsPricingDevelopers · APIContext Window · Glossary

Key facts · as of 2026-10-05

  • gpt-oss-20b by OpenAI. BenchGecko score 45.0, rank 171 of 312 scored models (normalized average of public benchmark scores).
  • List price $0.0180 input · $0.0900 output per 1M tokens (as of 2026-10-05).
  • Sold by 12 providers (as of 2026-10-05): Darkbloom (fp8) $0.0180 in / $0.0900 out · AkashML (fp4) $0.0200 in / $0.10 out · DekaLLM (bf16) $0.0290 in / $0.14 out · CoreWeave (fp4) $0.0300 in / $0.13 out · DeepInfra (bf16) $0.0300 in / $0.14 out · and 7 more. Every provider

How to cite · data as of 2026-10-05

gpt-oss-20b · benchmarks, pricing and providers. BenchGecko, data as of 2026-10-05. https://benchgecko.ai/model/gpt-oss-20b

Credit "Source: BenchGecko" with a link. Prices per provider and Gecko Tests are BenchGecko data (CC BY 4.0); benchmark scores keep their original source, listed in the JSON. JSON · llms.txt · MCP