The news: OpenAI is bringing its newest models to Amazon Web Services (AWS) for the first time, marking a major milestone in the ongoing battle for AI cloud dominance.

• Its open-weight models, gpt-oss-120b and gpt-oss-20b—available via Amazon Bedrock and SageMaker AI platforms—are able to handle complicated text-based operations and integrate into cloud-based systems.
• This move expands access to OpenAI models beyond Microsoft Azure, which has been its primary cloud partner until now.

The new OpenAI products are the company’s first "open" large language models (LLMs) since GPT-2 more than five years ago, per TechCrunch.

Opening up: This launch is a move to compete with open models from rivals like Meta’s Llama and Chinese challengers such as DeepSeek-R1 and Alibaba’s Qwen. In January, OpenAI CEO Sam Altman said the company has been “on the wrong side of history” when it comes to open sourcing models.

Unlike open-source models that generally include all architecture, training code, and data sets, OpenAI’s open-weight models only release final trained parameters.

Finer details: The main difference between the two new models is operating capacity: oss-120b can run on a single 80 GB GPU, while oss-20b can run on a standard home computer with 16 GB of memory.

Designed for reasoning tasks, they perform comparably to o-series models but have significantly higher rates of hallucination: oss-120b and oss-20b hallucinated in response to 49% and 53% of questions, per TechCrunch, compared with the o1 model’s 16%.

Why the shift? OpenAI is going platform-neutral, possibly to gain leverage in its tense relationship with Microsoft.

This surprise debut on AWS weakens Microsoft’s exclusivity as its primary cloud provider and puts pressure on Meta’s—now somewhat stale—open-source lead.

Amazon, facing pressure from Wall Street over slow ROI on AI, now gets a strategic advantage by beefing up its Bedrock and SageMaker offerings.

Wider access: More cloud AI flexibility means easier experimentation with OpenAI’s powerful models across more platforms, which could speed up go-to-market launches for AI-driven content, automation, and customer tools, especially for enterprise teams already using AWS.

Our take: OpenAI’s models are getting easier to access, meaning lower costs and fewer technical hurdles to trying powerful AI tools.

AWS customers should start testing oss-120b and oss-20b for things like generating subject lines, social copy, and campaign variations and explore ways to fine-tune the models with company data.