This website uses cookies to anonymously analyze website traffic using Google Analytics.
AcceptDecline

Models / OpenAI Open Weight

OpenAI

OpenAI Open Models. Deploy instantly.

Immediately start using gpt-oss-120B on Together AI's production-ready infrastructure with proven performance, high reliability and comprehensive developer tooling.

Try OpenAI Models
Join Technical Webinar

Deep Dive into OpenAI's Latest Models: Technical Webinar

Join us for an exclusive breakdown of how OpenAI's gpt-oss-120B and gpt-oss-20B actually work. Perfect for developers, researchers, and technical leaders who want to understand the architecture, training innovations, and practical deployment strategies.

Register Now

Deploy OpenAI's Open Models in Minutes

No infrastructure setup. No vendor lock-in. Complete freedom to fine-tune for specialized domains and build commercial applications with full model ownership.

# Install the Together AI library
pip install together

# Get started with OpenAI's open models
from together import Together
client = Together()

response = client.chat.completions.create(
    model="openai/gpt-oss-120B",
    messages=[{
        "role": "user", 
        "content": "Explain quantum computing with step-by-step reasoning"
    }],
    stream=True
)

for token in response:
    if hasattr(token, 'choices'):
        print(token.choices[0].delta.content, end='', flush=True)
View API Docs

Why OpenAI's Open-Weight Models Matter

Frontier AI with complete model ownership and deployment flexibility

‍

These aren't just scaled-up versions of existing models—they're purpose-built reasoning models released under Apache 2.0 licensing while maintaining the safety standards that enterprise deployment requires.

  • OPEN-WEIGHT MODEL RELEASE

    • OpenAI's gpt-oss-120B and gpt-oss-20B under Apache 2.0 license
    • Complete freedom to fine-tune for specialized domains, deploy anywhere, and build commercial applications
    • Apache 2.0 license allows unlimited modification and deployment globally

    ENABLING GLOBAL DEPLOYMENT

    • Enables deployment in emerging markets, resource-constrained sectors, and smaller organizations
    • Enables government digital sovereignty with air-gapped deployments
    • Provides frontier reasoning capabilities with complete deployment control

    Safety-first Approach

    • Comprehensive safety training and evaluation for each model
    • Rigorously tested maliciously fine-tuned versions under OpenAI's Preparedness Framework
    • Verified safety even after weight release to open community

OpenAI's Open Reasoning Models

Purpose-built reasoning models released under Apache 2.0 license for unlimited deployment flexibility.

gpt-oss-120B

Frontier Reasoning Model

Starting at $0.15/1M tokens

  • 120B

    Parameters

  • Apache 2.0

    License

  • Reasoning

    Optimized

  • MXFP4

    Quantization

Key Strengths:

  • Frontier reasoning performance

  • Complete deployment freedom

  • Enterprise-grade reasoning

  • Unlimited commercial use

Try it out

gpt-oss-20B

Efficient Reasoning Model

Starting at $0.05/1M tokens

  • 20B

    Parameters

  • Apache 2.0

    License

  • Cost

    Effective

  • High

    Throughput

Key Strengths:

  • Superior cost-performance ratio

  • Native tool use

  • Harmony response format

  • Fast inference

Try it out

Built for Democratic AI Infrastructure

OpenAI's open models represent breakthrough approaches to making frontier AI truly accessible and customizable.

  • Harmony Response Format

    Models trained on OpenAI's harmony response format for consistent, structured outputs. Enables full chain-of-thought access and improved reasoning transparency.

    Complete reasoning process visibility

  • Optimized Cloud Inference

    Native MXFP4 quantization and efficient architecture designed for high-throughput cloud deployment on Together AI's infrastructure.

    Superior cloud economics and performance

  • Apache 2.0 Licensing

    Permissive licensing allows unlimited modification, commercial deployment, and redistribution. No vendor dependencies or usage restrictions.

    Complete intellectual freedom

  • Enterprise-Grade Safety

    Comprehensive safety training and malicious fine-tuning testing under OpenAI's Preparedness Framework ensures production-ready deployment.

    Enterprise-grade safety assurance

  • Native Agentic Capabilities

    Built-in function calling, web browsing, Python code execution, and Structured Outputs. No additional training required for tool use.

    Ready-to-use agent functionality

Frontier Performance, Open Access

See how OpenAI's open models deliver competitive performance while maintaining complete deployment freedom.

Qwen3-235B-A22B

OpenAI Open Models vs Competition

Model

MMLU

GPQA Diamond

Humanity's Last Exam

AIME 2024

AIME 2025

gpt-oss-120B

90.0

80.1

19.0

96.6

97.9

gpt-oss-20b

85.3

71.5

17.3

96.0

98.7

OpenAI o3

93.4

83.3

24.9

95.2

98.4

OpenAI o4-mini

93.0

81.4

17.7

98.7

99.5

Benchmark results demonstrate competitive frontier performance with complete deployment freedom

Deploy on Together's Proven Infrastructure

Access OpenAI's open models through Together's optimized inference platform with enterprise-grade security and performance guarantees.

  • Serverless Endpoints

    Pay-per-token pricing with automatic scaling. No infrastructure setup, no throttling.

    Best for:

    • Getting started

    • Variable workloads

    • Prototyping

    • Cost optimization

    gpt-oss-120B:
    $0.15 input/1M tokens
    $0.60 output/1M tokens

    gpt-oss-20B:
    ‍$0.05 input/1M tokens
    $0.20 output/1M tokens

    Try Serverless

  • Dedicated Endpoints

    Guaranteed performance with custom fine-tuning capabilities. Complete control over model deployment.

    Best for:

    • Production applications

    • Custom fine-tuning

    • Predictable latency

    • Enterprise SLA

    gpt-oss-120B:
    $0.22/minute (4x H100)

    gpt-oss-20B:
    ‍$0.06/minute (1x H100)

    Request Deployment

  • Batch API

    Cost-effective processing for large workloads, distillation, and synthetic data generation.

    Best for:

    • Large-scale processing

    • Model distillation

    • Data generation

    • Research workloads

    50% discount on standard pricing

    Create Batch Job

Enterprise-Grade Security

Your data and models remain fully under your control with industry-leading security standards.

  • 99.9% Uptime SLA

    Multi-region deployment with proven reliability at scale

  • SOC 2 Type II Compliant

    Enterprise security standards with comprehensive auditing

  • Complete Model Ownership

    Deploy anywhere with full rights to model weights and fine-tuned versions

  • North American Infrastructure

    Data sovereignty with US-based secure server deployment

"We've been consistently impressed by Together's ability to quickly provide us with access to the latest open source models - often on day 1 of release - all while delivering exceptional performance and cost-efficiency."

‍

— Nick Huber, AI Ecosystem Lead, Poe

"Together AI's dedicated infrastructure delivers the enterprise reliability that The Washington Post requires for deploying AI applications into production. When breakthrough models like OpenAI's open source models become available, Together AI's platform ensures we can deploy them on day 1 with the performance standards our organization needs."

‍

— Sam Han, Chief AI Officer, The Washington Post

Frequently Asked Questions

How do these models differ from OpenAI's API models?

These are OpenAI's open-weight models, released under Apache 2.0 license. Unlike API models, you can download, modify, fine-tune, and deploy them anywhere without restrictions. You own the model weights completely.

What does Apache 2.0 licensing mean for my business?

Apache 2.0 is a permissive license allowing unlimited commercial use, modification, and redistribution. You can fine-tune for specialized domains, deploy on your infrastructure, and build commercial applications without licensing fees or usage restrictions.

Can I fine-tune these models for my specific use case?

Yes! Complete freedom to fine-tune for specialized domains. Use Together AI's Fine-Tuning API or download the weights to customize on your own infrastructure. All fine-tuned versions remain fully owned by you.

How do I migrate from OpenAI's API to these open models?

Migration is seamless with OpenAI-compatible API endpoints on Together AI. Simply change the model name in your existing code. You get the same reasoning capabilities with deployment freedom and predictable costs

What safety measures are included?

Each model underwent comprehensive safety training and evaluation. OpenAI rigorously tested maliciously fine-tuned versions under their Preparedness Framework to verify safety even after weight release to the community.

What are the current pricing rates?

gpt-oss-120B: $0.16 input / $0.60 output per million tokens and gpt-oss-20B: $0.05 input / $0.20 output per million tokens. Use our Batch API for 50% discounts on large workloads. No licensing fees ever.

Deploy OpenAI's Open-Weight Models

Access frontier reasoning models with complete deployment freedom. Join our technical webinar to understand these breakthrough models.

Try nowRegister for Technical Webinar

Subscribe to newsletter

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.