OpenAI Delays Open Source Model Indefinitely

Artificial intelligence start-up OpenAI has indefinitely delayed the release of an open-source model for “safety testing”, as the company struggles to address a wave of open-source competition initiated by China’s DeepSeek earlier this year.

The release was scheduled for this week following an earlier delay a month ago, but OpenAI chief executive Sam Altman said the company wanted to “review high-risk areas”.

“We need time to run additional safety tests and review high-risk areas,” Altman said on social media, adding that the company was “not yet sure” how long this would take.

OpenAI chief executive Sam Altman. Image credit: OpenAI

‘Get it right’

“Once weights are out, they can’t be pulled back. This is new for us and we want to get it right.”

Aidan Clark, OpenAI’s vice president of research who is leading the open model team, said on social media that the team needed “more time” to attain the high standards it’s aiming for.

“Our bar for an open-source model is high and we think we need some more time to make sure we’re releasing a model we’re proud of along every axis,” Clark wrote.

Unlike OpenAI’s closed-source models, the open-source offering would be available for modification by users, who could run it on local hardware.

The model is expected to have similar reasoning capabilities to OpenAI’s o-series models and OpenAI is reportedly aiming for it to top other open-source models on performance and capabilities.

Altman began speaking about the possibility of releasing an open model soon after DeepSeek made headlines with its low-cost, high-performance open-source models in January.

Open-weight model

In April he said the company would release its first open-weight model since GPT-2 in the coming months.

Open-weight models provide the ability to customise the weights, or relationships among the billions of parameters that are set during the model’s training, providing more visibility and control than closed systems.

Such models can generally be used on a company’s own hardware, meaning the model can handle a company’s sensitive data without it ever leaving the premises, a significant factor in areas such as banking or healthcare.

Chinese tech giants Alibaba and Baidu have since embraced an open-source approach, in a measure of the influence the small Hangzhou-based company has had on the industry.

Matthew Broersma

Matt Broersma is a long standing tech freelance, who has worked for Ziff-Davis, ZDnet and other leading publications

Recent Posts

Broadcom Upgrades Jericho Data Centre Chip For AI Age

Next-generation Jericho4 chip can link together data centre sites up to 100 km apart, bringing…

6 days ago

French AI Start-Up Mistral In Talks For $1bn Funding Round

Mistral reportedly seeks valuation of $10bn in funding round that includes venture capital groups and…

6 days ago

Amazon Breaks Up Wondery Podcast Business

Amazon to lay off 110 staff as it folds Wondery audio podcasts into Audible division…

6 days ago

Google Agrees To Cut AI Energy Use At Peak Times

Google cuts deals with two US power authorities to scale back AI data centre energy…

6 days ago

Crypto Exchange Bullish Seeks $4.23bn IPO Valuation

Crypto exchange Bullish seeks to go public on NYSE at valuation of up to $4.23bn,…

6 days ago

Former Huawei Engineers Jailed For Stealing Chip Tech

Former employees of Huawei unit HiSilicon sentenced to prison for stealing secrets, as China ramps…

6 days ago