OpenAI, the parent company of ChatGPT, is rolling out new open-weight generative artificial intelligence (AI) models as US tech giants attempt to gain an upper hand in the AI race. Here’s everything we know about OpenAI’s new models and how they compare with competing models from other companies.
OpenAI is releasing advanced open-weight reasoning models, its first open-weight models since the release of GPT-2 in 2019. The model comes in two variants, the 20-billion-parameter variant dubbed gpt-oss-20b and the 120-billion-parameter variant named gpt-oss-120b. However, it’s worth noting that the model is not open in a strict sense, and OpenAI does not share its key intellectual properties like proprietary architecture and training data.
The gpt-oss-20b is a medium-sized model that can run on most laptops and desktops. gpt-oss-120b is, however, a large model meant for data centers and high-end laptops, and desktops. According to OpenAI, the models are deeply customizable and are designed for agentic tasks.
The models are being released under the Apache 2.0 license, which means that they can be modified without any copyleft restrictions or patent risks. The models can be deployed through multiple sources like OpenRouter, Hugging Face, Databricks, Together.ai, Azure, Cloudflare, and AWS.
“We’re excited to make this model, the result of billions of dollars of research, available to the world to get AI into the hands of the most people possible,” said OpenAI CEO Sam Altman on the open-weight models.
Why Has OpenAI Launched Open-Weight Models?
Most US-based tech giants, with the notable exception of Meta Platforms, have opted for closed-source AI models. In OpenAI’s case, it was able to charge companies and developers for its AI models via an API.
However, as Chinese tech companies launched breakthrough open-sourced models in a flurry, OpenAI CEO Sam Altman admitted in January that the company had been “on the wrong side of history” by predominantly going with closed models. His comments came shortly after DeepSeek shook the world with its low-cost open-sourced AI model, which the company claims to have developed for just about $6 million – a fraction of what US tech companies spend on developing their models.
Through its open-weight models, OpenAI is targeting AI builders and developers, especially those looking for low-cost alternatives.
On a strategic level, open models would help in global adoption of US models, and the Trump administration has also called upon US tech giants to open-source more technology. In his statement to TechCrunch, Altman said, “Going back to when we started in 2015, OpenAI’s mission is to ensure AGI that benefits all of humanity.” He added, “To that end, we are excited for the world to be building on an open AI stack created in the United States, based on democratic values, available for free to all and for wide benefit.”
How do New OpenAI Models Compare with Others?
In its release, OpenAI did not share metrics comparing GPT-OSS to other open models like Google’s Gemma, Meta Platforms’ Llama, and open-source models from Chinese AI startup DeepSeek.
Meanwhile, the metrics shared by OpenAI (listed above) show that the performance of its open model is similar to its closed reasoning models on some benchmarks like American Invitational Mathematics Examination (AIME) 2025, Humanity’s Last Exam, and coding tasks.
How Safe are OpenAI’s New Models?
OpenAI has been delaying the launch of its open-weight models over safety issues. In its release, the company has tried to allay fears over the safety of its new models and said, “We rigorously tested a maliciously fine-tuned version of gpt-oss-120b under our Preparedness Framework, and found that it doesn’t reach high capability levels.”
It added, “These training and testing methods were reviewed and informed by external safety experts, and mark a meaningful advancement in open model safety standards.”
In 48 hours, OpenAI has leaked that they are at $12 billion or $13 billion annualized revenue, up from $10bn in June. When you actually plot these numbers out, their actual booked revenue is, at best, $5.25bn – if not slightly less. https://t.co/H5XGZcARpd pic.twitter.com/JnmXWpOtkI
— Ed Zitron (@edzitron) August 1, 2025
OpenAI’s Revenues Surge, But So Does Cash Burn
OpenAI’s revenues have been growing at a stellar pace, and according to The Information, its annualized revenue run rate was $12 billion, or $1 billion per month in July. For context, the annualized revenue run rate was $10 billion in June and $5.5 billion in December 2024, and the company looks on track to hit $20 billion by the end of this year.
OpenAI now boasts of 700 million weekly active users for ChatGPT products. For context, the number was just 400 million in February.
However, while the ChatGPT parent has more than doubled its annualized revenue run rate in just seven months this year, its losses and cash burn have also swelled over the same period. The company has increased its 2025 cash burn projection to $8 billion as compared to the $1 billion that it projected earlier in the year, as it continues to chase growth at the cost of the burgeoning cash burn.
For most AI startups, the surge in revenues has been accompanied by a more than commensurate rise in losses and cash burn. However, thanks to the generous support from existing and new investors, they have been able to not only stay afloat but also expand aggressively.
OpenAI, for instance, acquired Jony Ive’s startup, io Products, for $6.4 billion in May. Ive, who was one of the principal architects of the iPhone, founded the startup last year and has been partnering with OpenAI for two years.
OpenAI’s cash burn might only increase as the company forays into hardware post the acquisition of io Products.
AI Mania Is Back
Meanwhile, the cash burn hasn’t deterred investors’ optimism towards the company, and OpenAI is reportedly looking at a $500 billion valuation in an employee stock sale program.
While a few months back it seemed that the AI mania had subsided, US tech companies’ recent earnings calls show that the euphoria is back, and most leading tech companies are looking to further increase their capex towards AI. Meta is perhaps the most aggressive, and its CEO, Mark Zuckerberg, has spent billions hiring talent for the company’s Superintelligence Labs.
