OpenAI launches two ‘open' AI reasoning models
The models come in two sizes: a larger and more capable gpt-oss-120b model that can run on a single Nvidia GPU, and a lighter-weight gpt-oss-20b model that can run on a consumer laptop with 16GB of memory.
The launch marks OpenAI's first 'open' language model since GPT-2, which was released more than five years ago.
In a briefing, OpenAI said its open models will be capable of sending complex queries to AI models in the cloud, as TechCrunch previously reported. That means if OpenAI's open model is not capable of a certain task, such as processing an image, developers can connect the open model to one of the company's more capable closed models.
While OpenAI open-sourced AI models in its early days, the company has generally favored a proprietary, closed-source development approach. The latter strategy has helped OpenAI build a large business selling access to its AI models via an API to enterprises and developers.
However, CEO Sam Altman said in January he believes OpenAI has been 'on the wrong side of history' when it comes to open sourcing its technologies. The company today faces growing pressure from Chinese AI labs — including DeepSeek, Alibaba's Qwen, and Moonshot AI —which have developed several of the world's most capable and popular open models. (While Meta previously dominated the open AI space, the company's Llama AI models have fallen behind in the last year.)
In July, the Trump Administration also urged U.S. AI developers to open source more technology to promote global adoption of AI aligned with American values.
With the release of gpt-oss, OpenAI hopes to curry favor with developers and the Trump Administration alike, both of which have watched the Chinese AI labs rise to prominence in the open source space.
'Going back to when we started in 2015, OpenAI's mission is to ensure AGI that benefits all of humanity,' said OpenAI CEO Sam Altman in a statement shared with TechCrunch. '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 the models performed
OpenAI aimed to make its open model a leader among other open-weight AI models, and the company claims to have done just that.
On Codeforces (with tools), a competitive coding test, gpt-oss-120b and gpt-oss-20b score 2622 and 2516, respectively, outperformed DeepSeek's R1 while underperforming o3 and o4-mini.
On Humanity's Last Exam, a challenging test of crowd-sourced questions across a variety of subjects (with tools), gpt-oss-120b and gpt-oss-20b score 19% and 17.3%, respectively. Similarly, this underperforms o3 but outperforms leading open models from DeepSeek and Qwen.
Notably, OpenAI's open models hallucinate significantly more than its latest AI reasoning models, o3 and o4-mini.
Hallucinations have been getting more severe in OpenAI's latest AI reasoning models, and the company previously said it doesn't quite understand why. In a white paper, OpenAI says this is 'expected, as smaller models have less world knowledge than larger frontier models and tend to hallucinate more.'
OpenAI found that gpt-oss-120b and gpt-oss-20b hallucinated in response to 49% and 53% of questions on PersonQA, the company's in-house benchmark for measuring the accuracy of a model's knowledge about people. That's more than triple the hallucination rate of OpenAI's o1 model, which scored 16%, and higher than its o4-mini model, which scored 36%.
Training the new models
OpenAI says its open models were trained with similar processes to its proprietary models. The company says each open model leverages mixture-of-experts (MoE) to tap fewer parameters for any given question, making it run more efficiently. For gpt-oss-120b, which has 117 billion total parameters, OpenAI says the model only activates 5.1 billion parameters per token.
The company also says its open model was trained using high-compute reinforcement learning (RL) — a post-training process to teach AI models right from wrong in simulated environments using large clusters of Nvidia GPUs. This was also used to train OpenAI's o-series of models, and the open models have a similar chain-of-thought process in which they take additional time and computational resources to work through their answers.
As a result of the post-training process, OpenAI says its open AI models excel at powering AI agents, and are capable of calling tools such as web search or Python code execution as part of its chain-of-thought process. However, OpenAI says its open models are text-only, meaning they will not be able to process or generate images and audio like the company's other models.
OpenAI is releasing gpt-oss-120b and gpt-oss-20b under the Apache 2.0 license, which is generally considered one of the most permissive. This license will allow enterprises to monetize OpenAI's open models without having to pay or obtain permission from the company.
However, unlike fully open source offerings from AI labs like AI2, OpenAI says it will not be releasing the training data used to create its open models. This decision is not surprising given that several active lawsuits against AI model providers, including OpenAI, have alleged that these companies inappropriately trained their AI models on copyrighted works.
OpenAI delayed the release of its open models several times in recent months, partially to address safety concerns. Beyond the company's typical safety policies, OpenAI says in a white paper that it also investigated whether bad actors could fine-tune its gpt-oss models to be more helpful in cyber attacks or the creation of biological or chemical weapons.
After testing from OpenAI and third-party evaluators, the company says gpt-oss may marginally increase biological capabilities. However, it did not find evidence that these open models could reach its 'high capability' threshold for danger in these domains, even after fine-tuning.
While OpenAI's model appears to be state-of-the-art among open models, developers are eagerly awaiting the release of DeepSeek R2, its next AI reasoning model, as well as a new open model from Meta's new superintelligence lab.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
a minute ago
- Yahoo
US ethane curbs will make contracting to China harder, Energy Transfer says
By Arathy Somasekhar HOUSTON (Reuters) -Recent U.S. restrictions on ethane exports to China will likely make it more difficult to contract with Chinese companies, even though they have already been lifted, U.S. exporter Energy Transfer said on Wednesday. The U.S. placed restrictions on shipping ethane - and a wide swathe of other exports - to China in late May and early June after accusing Beijing of slowing shipments of rare earths vital to automakers and other industries. The restrictions were rescinded last month, but they disrupted flows of ethane and caused significant delays to shipments. "That, you know, put a little bit of a black eye on us, on our industry, on our country...," Marshall McCrea, co-CEO of Energy Transfer, said in a post-earnings conference call. The company is one of the top U.S. exporters of ethane, a natural gas liquid. "We think it's going to be probably a little bit more difficult to contract with Chinese crackers, good or bad, we think that they're probably going to be a little bit more hesitant," McCrea added. About half of U.S. ethane, which is extracted from shale gas, heads to China where it is run through crackers to produce ethylene, a building block for plastics. Chinese petrochemical firms use ethane as a feedstock because it is cheaper than naphtha, while U.S. oil and gas producers need China to buy their natural gas liquids as domestic supply exceeds demand. Rival Enterprise Products Partners also warned last week that the export curbs compromised the U.S. brand for reliable supply and energy security. "These kind of actions rarely hurt the intended target and often backfire hurting our own industry more," said Jim Teague, CEO of Enterprise Products. Enterprise said at least one non-Chinese company that it was in discussions with about contracting ethane decided to contract naphtha instead. Energy Transfer reported a 11.5% decline in net income to $1.16 billion, or 32 cents per unit, in the three months ended June 30. Revenue of $19.24 billion came in well below estimates of about $22 billion, according to LSEG data. Sign in to access your portfolio
New York Post
2 minutes ago
- New York Post
Hey, experts — admit what you got so wrong on Trump's tariffs
Economists across the political spectrum predicted that President Donald Trump's trade negotiations would end in disaster. Now that his Aug. 1 deadline has passed without the sky falling — and with multiple advantageous deals completed — it's time to seriously reevaluate the flawed arguments the experts made against his strategy. Many, it turns out, made basic errors in economic reasoning. Advertisement On the left, Nobel laureate and Columbia professor Joseph Stiglitz declared in January that Trump's policy was 'very bad for America and for the world,' while University of Michigan economist Justin Wolfers called it 'impressively destructive.' On the right, prominent free-market advocates like George Mason's Donald Boudreaux also voiced strong opposition. Advertisement Yet their arguments against tariffs revealed a fundamental misunderstanding: They decried tariffs as uniquely harmful, while ignoring that the same logic applies to all taxes. Take the common critique that tariffs, as a tax on trade, reduce trade overall. Phil Gramm and Larry Summers — one conservative, one liberal — jointly argued that tariffs 'distort domestic production' by pushing resources toward less efficient uses. They warned tariffs would slow economic growth. Advertisement That's true. But every tax, including sales taxes and income taxes, discourages trade, distorts production and reduces growth. Sales taxes lower consumption. Income taxes discourage work. Corporate taxes deter investment. All taxes distort the economy — tariffs are no exception. Advertisement Another frequent claim is that tariffs hurt consumers. Again, true — just as all taxes do. Logically, opposing tariffs simply because they raise prices and reduce growth means we should oppose all taxes. But unless we abolish government spending — which stands at $7 trillion this year — we need taxes of some kind. That's why economists usually argue for minimizing the total economic damage that all taxes cause across the board. Distortions increase as tax rates do. Before Trump's policies, the average US tariff rate stood at just 2.5% — tiny compared to the 43.4% average top personal income tax rate (including federal and state taxes) or the 27.5% average total corporate tax rate. If we understand a tariff as a tax like any other, higher tariffs could in fact reduce the overall economic burden on American individuals and companies — an outcome that Trump has often touted as his ultimate goal. Advertisement It's unclear whether a 15% tariff is optimal, but it seems apparent now that a 2.5% rate was too low. Economists also missed how negotiation tactics work. Trump began with aggressive tariff threats, horrifying many economists — but the results speak for themselves. The United States has secured deals that dramatically opened foreign markets representing 55% of global GDP. Advertisement Even critics have had to acknowledge the shift. 'To avoid worst of Trump tariffs, [the European Union] accepted a lopsided deal,' The Washington Post conceded, while the London-based Financial Times described how the EU 'succumbed to Trump's tariff steamroller.' 'Under the new deal, US goods into Vietnam will not be taxed while Vietnamese exports will face a 20% US tariff,' the South China Morning Post explained — in coverage that described Hanoi's 'optimism' regarding the agreement. So while the United States is imposing higher tariffs on many imports, other countries lowered or removed their tariffs on American goods, and dropped many of their non-tariff barriers as well. Advertisement These are significant wins that economists failed to anticipate, and that few thought remotely possible even six months ago. Experts also ignored yet another of Trump's reasons for increasing tariffs: as a means of providing for national defense and global freedom of the seas, costs that Americans have borne for a century. Ideally, other countries would help pay for these efforts — how about they just send us a check for the share of benefits they are receiving? Advertisement But since that's not about to happen, tariffs may be the only viable alternative. Trump's trade policies defied economists' dire predictions, delivering substantial gains in opening foreign markets to American exports without tanking the US economy. If tariffs can help lower more damaging taxes while advancing strategic national interests, they deserve a more honest and nuanced evaluation. At the very least, economists should have the guts to admit they were wrong — and take a hard look at their conventional wisdom. John R. Lott Jr., president of the Crime Prevention Research Center, is an economist who has held research or teaching positions at the University of Chicago, Wharton Business School, Stanford, Yale and UCLA.
New York Post
2 minutes ago
- New York Post
Trump announces 100% tariff on semiconductors made outside the US in potential blow to domestic businesses
WASHINGTON — President Trump announced plans Wednesday for a 100% tariff on 'all chips and semiconductors' — insisting that there would be an orderly economic transition despite potential major impacts on US businesses. 'We'll be putting a tariff, approximately 100%, on chips and semiconductors,' Trump said at an Oval Office event with Apple CEO Tim Cook, who arrived to announce plans for $600 billion in new investments in the US. Trump said he would exempt companies that have plans to transition their manufacturing to the US — a hint that the world's largest chip manufacturer, Taiwan's TSMC, which is building massive manufacturing plants in Arizona, would be spared. Advertisement 3 'We'll be putting a tariff approximately 100% on chips and semiconductors,' Trump said at an Oval Office event with Apple CEO Tim Cook. REUTERS 'If you're building in the United States of America, there's no charge, even though you're building and you're not producing yet in terms of the big numbers of jobs and all of the things that you're building. If you're building there will be no charge,' Trump said. 'So 100% tariff on all chips and semiconductors coming into the United States. But if you've made a commitment to build, or if you're in the process of building, as many are, there is no tariff.' Cook did not comment on the tariff plan when given an opportunity to address the press. Advertisement Trump described the duties as a way to hold accountable companies that backtrack on pledges to open US manufacturing plants. 3 Trump also said he would exempt companies that have plans to transition their manufacturing to the US. AFP via Getty Images 3 The president described his tariff plan as a way to hold accountable companies that backtrack on pledges. AFP via Getty Images 'If, for some reason, you say you're building and you don't build, then we go back and get we added up, it accumulates, and we charge you at a later date, you have to pay, and that's a guarantee,' Trump said. Advertisement 'I think the chip companies are all coming back home. They're all coming back. You know, we started with Intel, and gradually Intel was just taken over the coals, they were taken to the cleaners, frankly, and moved to other places in particular Taiwan, but I think a lot of those companies are coming back, and they're coming back very rapidly.' In additional to large 'reciprocal' tariffs and a new 10% baseline tariff on most countries, Trump previously applied 50% tariffs on foreign steel, aluminum and copper and a 25% tariff on foreign-made cars in an attempt to boost those American industries. The president did not specify a start date for the new computer-chip tariffs, and has previewed similar duties for pharmaceutical drugs and lumber.
