GPT-OSS-20B

v20250805

OpenAI

Modelopen-sourceapache-2.0self-hostedprivacy
92
Exceptional
About This Model

OpenAI's edge-optimized open-weight model released August 2025. 21B total params (3.6B active), Apache 2.0 license. Matches o3-mini despite small size. Runs in 16GB memory (edge devices).

Last Evaluated: November 17, 2025
Official Website

Trust Vector Analysis

Dimension Breakdown

🚀Performance & Reliability
+

Flagship open-source performance. MoE architecture activates 5.1B of 117B params per token. Matches or beats o4-mini on most benchmarks.

task accuracy code

Competition coding and tool use benchmarks

Evidence
Codeforces BenchmarkMatches o4-mini on competition coding
TauBench Tool CallingExceeds o4-mini on tool calling
highVerified: 2025-11-17
task accuracy reasoning

Math competition benchmarks

Evidence
AIME 2024 & 2025Outperforms o3-mini on competition mathematics
Chain-of-Thought AccessFull chain-of-thought reasoning process exposed
highVerified: 2025-11-17
task accuracy general

General knowledge and domain-specific testing

Evidence
MMLU & HLEMatches o4-mini on general problem solving
HealthBenchExceeds o4-mini on health-related queries
highVerified: 2025-11-17
output consistency

Internal testing

Evidence
OpenAI Model CardConfigurable reasoning effort for consistency
mediumVerified: 2025-11-17
latency p50

Median latency estimation

Evidence
Optimized InferenceFast inference with MoE architecture
mediumVerified: 2025-11-17
latency p95

95th percentile from community benchmarks

Evidence
Community Deployments~2s on H100 hardware
mediumVerified: 2025-11-17
context window

Official specification

Evidence
OpenAI Technical Specs128K context window natively supported
highVerified: 2025-11-17
uptime

Self-hosting provides full control

Evidence
Self-Hosted Model100% uptime when self-hosted
highVerified: 2025-11-17
🛡️Security
+

Good base security. Self-hosting provides complete control over safety guardrails and data handling.

prompt injection resistance

OWASP LLM01 testing

Evidence
OpenAI Safety TestingGood resistance, customizable for self-hosted
mediumVerified: 2025-11-17
jailbreak resistance

Adversarial testing

Evidence
Community TestingStandard resistance, self-host allows custom guardrails
mediumVerified: 2025-11-17
data leakage prevention

Self-hosting analysis

Evidence
Self-Hosted DeploymentComplete data control when self-hosted
highVerified: 2025-11-17
output safety

Safety testing

Evidence
OpenAI SafetyStandard safety training, customizable
mediumVerified: 2025-11-17
api security

Deployment security review

Evidence
Self-Hosted SecurityCustomer controls all API security when self-hosted
highVerified: 2025-11-17
🔒Privacy & Compliance
+

Perfect privacy when self-hosted. No data sent to OpenAI. Full compliance control. Ideal for regulated industries.

data residency

Self-hosting analysis

Evidence
Open-Weight ModelDeploy anywhere, full data residency control
highVerified: 2025-11-17
training data optout

Privacy model analysis

Evidence
Self-Hosted ModelNo data sent to OpenAI when self-hosted
highVerified: 2025-11-17
data retention

Self-hosting review

Evidence
Self-Hosted DeploymentComplete control over data retention
highVerified: 2025-11-17
pii handling

Data flow analysis

Evidence
On-Premises DeploymentPII never leaves your infrastructure
highVerified: 2025-11-17
compliance certifications

Compliance model review

Evidence
Self-Hosted ComplianceInherit your infrastructure's certifications
highVerified: 2025-11-17
zero data retention

Privacy architecture review

Evidence
Open-Weight ModelComplete control, zero external retention
highVerified: 2025-11-17
👁️Trust & Transparency
+

Exceptional transparency. Full chain-of-thought access. Complete model weights and architecture disclosed. Open-source enables auditing.

explainability

Reasoning transparency

Evidence
Full Chain-of-ThoughtComplete access to reasoning process
highVerified: 2025-11-17
hallucination rate

QA testing

Evidence
Benchmark TestingGood factual accuracy
mediumVerified: 2025-11-17
bias fairness

Bias benchmarks

Evidence
OpenAI Model CardStandard bias testing
mediumVerified: 2025-11-17
uncertainty quantification

Confidence assessment

Evidence
Model BehaviorGood uncertainty expression
mediumVerified: 2025-11-17
model card quality

Documentation review

Evidence
Comprehensive Model CardDetailed technical specs, benchmarks, architecture
highVerified: 2025-11-17
training data transparency

Training data disclosure review

Evidence
OpenAI DocumentationMostly English, STEM, coding focus disclosed
highVerified: 2025-11-17
guardrails

Safety mechanism review

Evidence
Customizable GuardrailsStandard safety, customizable when self-hosted
highVerified: 2025-11-17
⚙️Operational Excellence
+

Exceptional operational flexibility. Apache 2.0 enables commercial use. Massive deployment ecosystem. Self-host or use managed platforms.

api design quality

API compatibility review

Evidence
Deployment PlatformsWorks with vLLM, Ollama, llama.cpp, Azure, AWS, etc.
highVerified: 2025-11-17
sdk quality

SDK ecosystem review

Evidence
GitHub RepositoryOfficial repo, Hugging Face integration
highVerified: 2025-11-17
versioning policy

Version stability analysis

Evidence
Open WeightsWeights frozen, no deprecation risk
highVerified: 2025-11-17
monitoring observability

Monitoring capability review

Evidence
Self-Hosted ControlFull observability when self-hosted
highVerified: 2025-11-17
support quality

Support ecosystem assessment

Evidence
Community SupportGitHub issues, community forums, deployment partners
mediumVerified: 2025-11-17
ecosystem maturity

Ecosystem breadth analysis

Evidence
Deployment PartnersAzure, Hugging Face, AWS, Fireworks, Together AI, Databricks, Vercel, Cloudflare, OpenRouter
highVerified: 2025-11-17
license terms

License review

Evidence
Apache 2.0 LicensePermissive Apache 2.0, no copyleft, no patent risk
highVerified: 2025-11-17
Strengths
  • +Apache 2.0 open-weight license enables commercial use without restrictions
  • +Matches o3-mini performance despite small 21B size (3.6B active)
  • +Runs in only 16GB memory (edge devices, consumer GPUs, IoT deployment)
  • +Complete data privacy when self-hosted (zero external data transmission)
  • +Ultra-low infrastructure costs (~$0.50-1/hr, 1/4 cost of 120B)
  • +Full chain-of-thought access and massive deployment ecosystem
Limitations
  • !Smaller capacity than gpt-oss-120b for complex tasks
  • !Self-hosting complexity and infrastructure costs
  • !Community support vs enterprise SLA
  • !Slightly lower performance than flagship closed models
  • !No built-in safety guardrails (customizable but requires setup)
Metadata
pricing
input: Free (self-hosted)
output: Free (self-hosted)
notes: Infrastructure costs only: ~$0.50-1/hr for consumer GPUs. Can run on edge devices. Free for download and commercial use under Apache 2.0.
last verified: 2025-11-17
context window: 128000
languages
0: English
1: Spanish
2: French
3: German
4: Italian
5: Portuguese
6: Japanese
7: Korean
8: Chinese
modalities
0: text
api endpoint: Self-hosted (various platforms)
model download: https://huggingface.co/openai/gpt-oss-20b
github: https://github.com/openai/gpt-oss
open source: true
license: Apache 2.0
architecture: Mixture-of-Experts (MoE) Transformer
parameters: 21B total (3.6B active per token)
memory requirement: 16GB (edge-optimized)
tokenizer: o200k_harmony
deployment platforms
0: Azure
1: AWS
2: Hugging Face
3: vLLM
4: Ollama
5: llama.cpp
6: LM Studio
7: Fireworks
8: Together AI
9: Baseten
10: Databricks
11: Vercel
12: Cloudflare
13: OpenRouter

Use Case Ratings

code generation

Excellent coding. Matches o4-mini. Configurable reasoning effort. Full chain-of-thought debugging.

customer support

Good for customer support. Self-host for complete data privacy. Configurable reasoning for cost control.

content creation

Strong content creation. Self-hosting enables unlimited generation without API costs.

data analysis

Excellent for data analysis. Keep sensitive data on-premises. Full chain-of-thought for transparency.

research assistant

Outstanding for research. 128K context. Self-host proprietary research data. Full reasoning transparency.

legal compliance

Perfect for legal. Self-host for complete compliance. No data leaves premises. Apache 2.0 license clarity.

healthcare

Ideal for healthcare. Self-host for HIPAA. Complete PHI privacy. No external data transmission.

financial analysis

Excellent for finance. Outperforms o3-mini on math. Self-host proprietary financial data.

education

Great for education. Full chain-of-thought shows reasoning steps. Self-host for institutional control.

creative writing

Good creative writing. Unlimited generation when self-hosted. No API costs for iteration.