New Open Models Openai are available at the Azure AI foundry
The new Open Open models are now available via Microsoft Azure AI Foundry and Windows AI Foundry. This means that developers can now build, test and deploy GPT-OS-20B and GPT-OS-15B with platforms for artificial intelligence.
The launch of AI Sam Altman launched its first “Open Weight” Sinceday 2019 models on Tuesday. The smaller of the two, GPT-OS-20B, has 21 billion parameters and is similar to O3-mini. It only needs 16 GB VRAM, so it can be launched at the rage of Windows hardware, including GPU with sufficient capacity.
Larger, GPT-OS-120B, has 117 billion parameters, can run on one NVIDIA GPU and is comparable in performance with O4-mini. Offered as an option without a server on the Azure Cloud infrastructure, prices for this model start at $ 0.15 per million input tokens and $ 0.60 per million output tokens.
The GPT-OS-20B uses managed calculation and its prices depend on the type of Azure Machine Learning virtual machine rather than processing tokens.
In addition to the Azure AI foundry, GPT – OSS – 20b is located locally via Windows AI Foundry for Windows 11, while MacOS support comes soon through Foundry Local. Both models are also available on more than a dozen other third -party platforms, including hugging the face, cloudflare and AWS.
What can the new AI models do and how to do it
Openai says that GPT-OS models are able to perform advanced thinking tasks, write code, search a site, and create a caroomous that works on behalf of the user; However, they have been trained, we only have a data file for text, so they will not be able to process or generate images, sound, or other media types.
Both Openai models use the “mixture of experts” to improve the efficient model, while the GPT-OS-120B activates only 5.1 billion parameters per token. They were trained using strengthening learning, where they read interaction with a simulated wheel and receiving feedback in the form of rewards to improve their future decisions.
Developers are free to adapt open models on Azure Ai Foundry
The GPT-OS models were licenses within Apache 2.0, allowing free use, modification and redistribution, for commercial purposes. Microsoft says that Azure AI Foundry offers tools for fast model training, weight management and low latency deployment. Own versions or checkpoints can be supplied in a few hours than weeks.
Azure AI Foundry tools allow developers to adapt GPT-OS models in different ways, included Splick into profitical data, optimize them for marginal devices, and export them to use canned ships such as Azure Kubernetes or Foundry Local.
“Whether you adapt to the domain -specific dome, you compress offline or prototyping before scaling, Azure AI Foundry and Foundry Local gives you a tool to do it all,” Microsoft said in his press release.
Both models will soon be compatible with the API Openai API reaction, which allows developers to confuse them in existing applications with minimal code changes.
Microsoft collects AI partnership for its Azure AI foundry
In May, Microsoft added Elona Musk Grok to Azure AI Foundry, a controversial step with regard to its narrow partnership with Open and Musk’s Nail -ing Feud with the company.
It also has non -exclusive licensing agreements with inflection AI and Mistral AI and Azure AI Foundry features of the foundation models of similar meta, Deecseek, Nvidia, Cohere and hugging the face. Drawn by the selection of the model, the more attractive the platform becomes for business and developers.
The voltage between Open and Microsoft has recently escalated because both companies are negotiating the conditions of their long -term partnership.