OpenAI, the dad or mum firm of ChatGPT, is rolling out new open-weight generative synthetic intelligence (AI) fashions as US tech giants try to realize an higher hand within the AI race. Right here’s every thing we learn about OpenAI’s new fashions and the way they evaluate with competing fashions from different corporations.
OpenAI is releasing superior open-weight reasoning fashions, its first open-weight fashions because the launch of GPT-2 in 2019. The mannequin is available in two variants, the 20-billion-parameter variant dubbed gpt-oss-20b and the 120-billion-parameter variant named gpt-oss-120b. Nevertheless, it’s price noting that the mannequin is just not open in a strict sense, and OpenAI doesn’t share its key mental properties like proprietary structure and coaching knowledge.
The gpt-oss-20b is a medium-sized mannequin that may run on most laptops and desktops. gpt-oss-120b is, nevertheless, a big mannequin meant for knowledge facilities and high-end laptops, and desktops. In line with OpenAI, the fashions are deeply customizable and are designed for agentic duties.
The fashions are being launched underneath the Apache 2.0 license, which implies that they are often modified with none copyleft restrictions or patent dangers. The fashions could be deployed by a number of sources like OpenRouter, Hugging Face, Databricks, Collectively.ai, Azure, Cloudflare, and AWS.
“We’re excited to make this mannequin, the results of billions of {dollars} of analysis, accessible to the world to get AI into the palms of the most individuals doable,” mentioned OpenAI CEO Sam Altman on the open-weight fashions.
Why Has OpenAI Launched Open-Weight Fashions?
Most US-based tech giants, with the notable exception of Meta Platforms, have opted for closed-source AI fashions. In OpenAI’s case, it was capable of cost corporations and builders for its AI fashions through an API.
Nevertheless, as Chinese language tech corporations launched breakthrough open-sourced fashions in a flurry, OpenAI CEO Sam Altman admitted in January that the corporate had been “on the incorrect aspect of historical past” by predominantly going with closed fashions. His feedback got here shortly after DeepSeek shook the world with its low-cost open-sourced AI mannequin, which the corporate claims to have developed for almost $6 million – a fraction of what US tech corporations spend on growing their fashions.
Via its open-weight fashions, OpenAI is focusing on AI builders and builders, particularly these on the lookout for low-cost alternate options.
On a strategic degree, open fashions would assist in international adoption of US fashions, and the Trump administration has additionally known as upon US tech giants to open-source extra know-how. In his assertion to TechCrunch, Altman mentioned, “Going again to after we began in 2015, OpenAI’s mission is to make sure AGI that advantages all of humanity.” He added, “To that finish, we’re excited for the world to be constructing on an open AI stack created in the US, primarily based on democratic values, accessible without spending a dime to all and for broad profit.”
How do New OpenAI Fashions Examine with Others?
In its launch, OpenAI didn’t share metrics evaluating GPT-OSS to different open fashions like Google’s Gemma, Meta Platforms’ Llama, and open-source fashions from Chinese language AI startup DeepSeek.
Supply
In the meantime, the metrics shared by OpenAI (listed above) present that the efficiency of its open mannequin is just like its closed reasoning fashions on some benchmarks like American Invitational Arithmetic Examination (AIME) 2025, Humanity’s Final Examination, and coding duties.
How Secure are OpenAI’s New Fashions?
OpenAI has been delaying the launch of its open-weight fashions over issues of safety. In its launch, the corporate has tried to allay fears over the security of its new fashions and mentioned, “We rigorously examined a maliciously fine-tuned model of gpt-oss-120b underneath our Preparedness Framework, and located that it doesn’t attain excessive functionality ranges.”
It added, “These coaching and testing strategies had been reviewed and knowledgeable by exterior security specialists, and mark a significant development in open mannequin security requirements.”
In 48 hours, OpenAI has leaked that they’re at $12 billion or $13 billion annualized income, up from $10bn in June. If you really plot these numbers out, their precise booked income is, at greatest, $5.25bn – if not barely much less. https://t.co/H5XGZcARpd pic.twitter.com/JnmXWpOtkI
— Ed Zitron (@edzitron) August 1, 2025
OpenAI’s Revenues Surge, However So Does Money Burn
OpenAI’s revenues have been rising at a stellar tempo, and in response to The Data, its annualized income run fee was $12 billion, or $1 billion monthly in July. For context, the annualized income run fee was $10 billion in June and $5.5 billion in December 2024, and the corporate seems on observe to hit $20 billion by the tip of this 12 months.
OpenAI now boasts of 700 million weekly energetic customers for ChatGPT merchandise. For context, the quantity was simply 400 million in February.
Nevertheless, whereas the ChatGPT dad or mum has greater than doubled its annualized income run fee in simply seven months this 12 months, its losses and money burn have additionally swelled over the identical interval. The corporate has elevated its 2025 money burn projection to $8 billion as in comparison with the $1 billion that it projected earlier within the 12 months, because it continues to chase development at the price of the burgeoning money burn.
For many AI startups, the surge in revenues has been accompanied by a greater than commensurate rise in losses and money burn. Nevertheless, because of the beneficiant assist from current and new traders, they’ve been capable of not solely keep afloat but in addition develop aggressively.
OpenAI, for example, acquired Jony Ive’s startup, io Merchandise, for $6.4 billion in Could. Ive, who was one of many principal architects of the iPhone, based the startup final 12 months and has been partnering with OpenAI for 2 years.
OpenAI’s money burn may solely improve as the corporate forays into {hardware} put up the acquisition of io Merchandise.
AI Mania Is Again
In the meantime, the money burn hasn’t deterred traders’ optimism in the direction of the corporate, and OpenAI is reportedly taking a look at a $500 billion valuation in an worker inventory sale program.
Whereas a number of months again it appeared that the AI mania had subsided, US tech corporations’ current earnings calls present that the euphoria is again, and most main tech corporations wish to additional improve their capex in the direction of AI. Meta is maybe probably the most aggressive, and its CEO, Mark Zuckerberg, has spent billions hiring expertise for the corporate’s Superintelligence Labs.
