$450 and 19 hours is all it takes to rival OpenAI’s o1-preview

Open-source approaches continue to show promise in democratizing artificial intelligence (AI).

NovaSky’s Sky-T1-32B-Preview

On Friday, the NovaSky research team at UC Berkeley released a new reasoning model, Sky-T1-32B-Preview, that performs comparably to OpenAI’s o1-preview — only it’s open source and was built in just 19 hours for under $450 using eight Nvidia H100 GPUs.

Also: The best open-source AI models: All your free-to-use options explained

The team developed Sky-T1 by fine-tuning Alibaba’s Qwen2.5-32-Instruct and trained it on data generated with QwQ-32B-Preview, another open-source model comparable to o1-preview. Using synthetic training data can help lower costs.

“We curate the data mixture to cover diverse domains that require reasoning, and a reject sampling procedure to improve the data quality. We then rewrite QwQ traces with GPT-4o-mini into a well-formatted version, inspired by Still-2, to improve data quality and ease parsing,” the team says of their data preparation process in the blog.

Outperforming OpenAI’s o1-preview

The model performed at or above o1-preview’s level on math and coding benchmarks but did not surpass o1 on the graduate-level benchmark GPQA-Diamond, which includes more advanced physics-related questions. NovaSky open-sourced all parts of the model, including weights, data, infrastructure, and technical details.

Also: OpenAI’s o1 lies more than any major AI model. Why that matters

o1 is now out of preview and is therefore more capable than its initial release. Plus, OpenAI is already preparing to launch o3, which the company says can outperform o1. But as the NovaSky team points out in their blog, the fact that Sky-T1 could be built so quickly still “demonstrate[es] that it is possible to replicate high-level reasoning capabilities affordably and efficiently.”

A more affordable reasoning model

The relatively short 19-hour training time means Sky-T1 cost just $450 to build, according to Lambda Cloud pricing, the team clarifies in the blog post. Considering GPT-4 used a suspected $78 million in compute, it is no small feat to present an example of a more affordable reasoning model that can be replicated by academic and open-source groups that lack OpenAI’s funding.

Almost half of those adopting generative AI want it to be open-source, citing cost and trust concerns. Continued breakthroughs in open-source AI could create a more even playing field for smaller labs, nonprofits, and other entities to develop competitive models — a refreshing turn for a new field already dominated by tech giants.

Keep your entertainment at your fingertips with the Amazon Fire TV Stick 4K! Enjoy streaming in 4K Ultra HD with access to top services like Netflix, Prime Video, Disney+, and more. With an easy-to-use interface and voice remote, it’s the ultimate streaming device, now at only $21.99 — that’s 56% off!

With a 4.7/5-star rating from 43,582 reviews and 10K+ bought in the past month, it’s a top choice for home entertainment! Buy Now for $21.99 on Amazon!