Big news in the AI world: OpenAI has just released its first open-weight language models since GPT2. Called GPT-OSS, these models are fully open, permissively licensed under Apache 2.0, and come in two sizes—120B and 20B parameters. Dell Technologies is the first and only OEM making these available to enterprises through the Dell Enterprise Hub on Hugging Face.
Why does this matter? For enterprises, it means you can finally run cutting-edge reasoning models entirely on-premises, customize them as needed, and integrate them tightly with your existing data systems—no vendor lock-in, no data leaving your firewall. And when you pair these models with the Dell AI Factory, you get a complete, validated path to enterprise-grade AI deployment.
What’s new with GPT‑OSS?
The GPT-OSS release is significant for a few reasons:
- Two model sizes for different needs: The 120B version delivers state-of-the-art reasoning, while the 20B model is light enough to run on a single GPU, or even a Dell AI PC.
- Permissive licensing (Apache 2.0): You can host, modify, finetune and redistribute these models—ideal for enterprise teams who need flexibility and control.
- Text only and reasoning-optimized: Perfect for things like chatbots, document analysis, retrieval augmented generation (RAG) and internal knowledge agents.
This is OpenAI signaling that open models aren’t just for researchers anymore—they’re now powerful enough for enterprise workloads.
How are these models built?
Under the hood, GPT-OSS models use a few design choices that make them particularly appealing for on-premises deployments:
- Instead of activating every parameter for every token, the Mixture of Experts (MoE) architecture activates only a subset, which keeps inference fast and cost-efficient.
- 4bit quantization (MXFP4) shrinks memory usage dramatically, allowing large models to run on fewer GPUs, maximizing efficiency.
- Benchmark scores put these models in the same league as OpenAI’s proprietary o-series models like o3-mini and o4-mini, elevating competitive reasoning performance.
Put simply, they’re powerful, efficient and open—three things enterprises have been asking for.
Why enterprises should care
As enterprises turn to AI, data security, compliance and cost considerations are top of mind to ensure responsible implementation. Open-weight models like GPT-OSS address this head on:
- Users can deploy them entirely on-premises, behind a firewall, with no external API calls.
- Enterprises get full control over fine-tuning, auditing and extending the model to their domain.
- Recurring API fees are avoided, which can add up quickly at scale.
This release lets organizations align AI strategy with their security policies and budget realities without compromising on capability.
Enter Dell AI Factory
The Dell AI Factory provides everything enterprises need to implement AI: Dell PowerEdge servers, GPU acceleration, storage, networking, orchestration software, AI PCs and managed services. When you drop GPT-OSS into this stack, you get a ready-made path to on-prem AI that’s validated end-to-end.
Some highlights:
- Hardware that matches the models: Dell XE-series PowerEdge servers (like XE9780 and XE9780L) are optimized for high-end GPUs including NVIDIA and AMD—the fuel GPT-OSS needs for fast inference and training.
- Scalable from edge to core: The 20B model runs comfortably on smaller nodes for local teams including on AI PCs, while the 120B model powers large-scale workloads in central data centers.
- Cost and efficiency: Dell estimates up to 62% lower costs for LLM inference compared to public cloud deployments. Pair that with the lack of API fees for open source models and the savings are significant.
Security and compliance, built-in
Both OpenAI and Dell have invested heavily in making this combination enterprise-ready:
- GPT-OSS models went through extensive safety reviews and red-team testing before release.
- Dell AI Factory layers in Zero Trust architecture, encryption and compliance certifications (HIPAA, FedRAMP, etc.) for highly regulated industries.
- And most importantly, the data never leaves the data center.
For industries like finance, healthcare or government, this is exactly the kind of set-up that makes secure AI adoption feasible.
Flexibility and ecosystem choice
Open-weights mean you’re not locked into a single vendor or roadmap. This allows users to fine-tune the models with their own data and build RAG pipelines using internal knowledge bases. Enterprises can also integrate with platforms like the Dell Enterprise Hub on Hugging Face or NVIDIA NeMo for orchestration and monitoring.
This flexibility means you can innovate at your own pace, and adapt as your needs evolve.
Why this matters now
The combination of OpenAI’s open models and the Dell AI Factory marks a shift towards powerful reasoning models that are now accessible, customizable and deployable on your terms.
For enterprises looking for AI that is:
- Secure, running behind a firewall,
- Cost-effective with no per-token API fees,
- Customizable with the ability to fine-tune to your domain, and
- High performance featuring state-of-the-art reasoning…
…the wait is over.
GPT-OSS plus the Dell AI Factory gives enterprises the best of both worlds—the latest AI capabilities with the control and security of on-prem deployment. It’s a new chapter for enterprise AI, and it’s one that puts enterprises firmly in the driver’s seat.
Check out the Dell Enterprise Hub on Hugging Face at https://dell.huggingface.co/.
