OpenAI has released the weights for two new models: gpt-oss-120b and gpt-oss-20b, marking a pivotal milestone for the AI community and commercial deployment.
We’re proud to announce that our platform is among the first to support these models in production, enabling customers to integrate them immediately into existing workflows — with or without OpenAI's API.
Overview of the Models
gpt-oss-120b
- 120 billion parameters
- Designed for advanced reasoning, planning, and agentic applications
- Suitable for data centers and high-end desktop systems
gpt-oss-20b
- 20 billion parameters
- Lightweight model optimized for most desktops and laptops
Shared Capabilities
- Instruction following and chain-of-thought reasoning
- Native support for Python code execution
- Full-parameter fine-tuning for domain-specific tasks
- Safety training aligned with OpenAI’s Preparedness Framework
- Permissively licensed for commercial use
- Available for download on Hugging Face and GitHub
Performance Benchmarks
- MMLU: 90.0 (120B), 85.3 (20B)
- GPQA Diamond: 80.1 (120B), 71.5 (20B)
- AIME 2024/2025: 96.6 / 97.9 (120B), 96.0 / 98.7 (20B)
What This Means for You
These open models allow you to run OpenAI-level LLMs entirely on your own infrastructure, while benefiting from our deployment orchestration and integration layers.
Deploy OpenAI Models Without an API
You can now self-host gpt-oss models on:
- Bolder-managed infrastructure (pre-optimized for these models)
- Your own cloud or on-prem servers
- Edge and mobile environments (20B variant)
With this shift, customers gain:
- Cost efficiency by eliminating usage-based API billing
- Full control over latency, data residency, and compliance
- Flexibility to fine-tune and modify models for specific domains
Seamless Integration with Existing Products
Whether you’re using our copilots, RAG-based applications, or Bolder Agents, these models can be swapped in with minimal to no code changes. Our runtime supports:
- Model abstraction via config-based switching
- Centralized observability and logging
- Backward compatibility with API- and weights-based models
Bolder Agents: Optimized for gpt-oss
Bolder Agents, our orchestration framework for AI-powered automation, now supports both gpt-oss models out of the box.
Capabilities include:
- Multi-step reasoning with constraint handling
- High-recall retrieval over enterprise datasets
- Domain-specific memory and context chaining
- Private and secure deployment, with optional tool integrations
These models offer significantly improved planning and execution accuracy for use cases ranging from document analysis to customer support automation.
Getting Started
If you're an existing customer, you can start using these models today:
-
Select your preferred deployment mode:
- Continue using OpenAI API
- Switch to self-hosted gpt-oss models
- Deploy gpt-oss within your existing infrastructure
-
Update your runtime configuration to enable the new models
-
Contact our solutions team for fine-tuning support or infrastructure setup
Documentation and reference deployments are now available in your dashboard.
Why This Matters
OpenAI’s release of these weights changes how AI can be built, deployed, and trusted. Instead of being tied to proprietary APIs, organizations now have direct access to performant models with transparent licensing and full tuning flexibility.
We believe this marks the beginning of a more open, interoperable era in applied AI — and we are proud to help lead that shift.
Final Thoughts
The gpt-oss models are ready for real-world use: they are fast, accurate, safety-evaluated, and flexible enough to support both research and enterprise deployments.
Whether you're building AI copilots, document agents, or internal LLM tooling, our platform gives you the tools to deploy these models securely, reliably, and efficiently.
Need help transitioning to a self-hosted model strategy? Contact us — we’re here to guide you every step of the way.