Openai, developer of ChatGPT, has released two major language models (LLMS) under the Apache 2.0 open source license. The models GPT-OSS-120B and GPT-OSS-20B are open-weight language models, claiming Openai offers powerful real-world performance at a low cost.
According to Openai, the new model is superior to similarly sized open models for inference tasks, and is optimized for efficient deployment of consumer hardware.
Openai said the GPT-OSS-120B model achieves prominent characteristics with Openai O4-Mini with core inference benchmarks, while efficiently running on a single 80GB GPU. The GPT-OSS-20B model has similar results for the Openai O3-Mini in a general benchmark, saying it can run on edge devices with just 16 GB of memory.
Graphics Processor Unit (GPU) manufacturer Nvidia says Openai's new model will be trained on an NVIDIA H100 GPU and can use NVIDIA NIM microservices.
Nvidia said that with software optimization on the NVIDIA Blackwell platform, the model can achieve 1.5 million tokens per 1.5 million tokens when run on an NVIDIA GB200 NVL72 system to support AI inference.
“We're looking forward to seeing you in the future,” said Amanda Brock, CEO of Openuk. “The beauty of AI's open source and openness is to nurture all aspects of the global discussion needs. It has the power to create digital public good access for everyone, but it is commercially accessible as well as beloved software for large scale technology. It enables the global reach and impact of AI.”
The main advantage of the open AI model is that it is not closed. This means that anyone can check it. This helps to improve the quality, remove bugs, and move on in some way to tackle bias when the source data the model is being trained is not diverse enough. The open model provides businesses with a way to fine-tune LLM to how their organization runs. However, CIOs face significant operational costs associated with deploying AI models, particularly so they need to weigh the benefits of using their own open AI models.
Gartner's senior director and analyst Haritha Khandabattu said Open Models, popularized by LLMSs such as Meta's Llama, are primarily used in the regulated industry. “These industries tend to experiment with open models,” she said. “Depending on where and how your open model is deployed, critical infrastructure may also be required.”
Khandabattu said that the reason organizations are experimenting with open models is to maintain control. However, from the IT leader she spoke to, Khandabattu said the total cost of the deployment was “very high.” There are significant operating costs and engineering costs required to customize, run and maintain open models.
She added that open models used in AI applications such as AI-based coding do not always match the performance of their proprietary models. She said this could negatively affect the organization, such as a decline in employee overall experience and developer experience and slow operational performance.
Khandabattu urged IT leaders to consider the pros and cons of open models. This could provide the corporate level of support your organization needs. “Like enterprise open source software, they also have their own risks,” she added.
