Latest news with #GPTOSS120B
Yahoo
4 days ago
- Business
- Yahoo
OpenAI launches two new AI models ahead of GPT-5 - here's everything you need to know
When you buy through links on our articles, Future and its syndication partners may earn a commission. OpenAI is once again doing side quests in the lead up to the launch of GPT-5. As we wait for the big update, OpenAI is pausing to bring us not one, but two entirely separate models to play with. Both of these new models are available to download for free to anyone with some coding ability via Hugging Face. They come in two sizes, with the larger option being the more capable gpt-oss-120b model that can run on just one single Nvidia GPU, and a second smaller model, called gpt-oss-20b. This one can run on a consumer laptop with 16GB of memory. This is the first time OpenAI has launched an open weight model in years, and has been delaying its release for a while now. While smaller AI companies like Le Chat, Deepseek, and Alibaba have frequently released open-weight models, OpenAI has tended to keep their doors closed off. Sam Altman, CEO of OpenAI, said at the start of the year that OpenAI felt it was on the wrong side of history for this, suggesting they would be going back to launching some open-source model What are open-weight models? Quite simply, an open-weight model is one where all of its training parameters are made publicly available. Developers can access these, analyzing and fine tuning them for their own projects. In such a competitive market, it seems strange for this to be a thing. And yet, it is a very popular option, with some of the most powerful models on the market being open-weighted. Of course, GPT-5 won't be, neither would the likes of Grok and Claude's top models. But that isn't to say that this new option from OpenAI isn't powerful. When put through tests, OpenAI's two new models both performed ahead of Deepseek's R1 and in a similar line to some of OpenAI's other reasoning models. In both models, the full chain of thought can be accessed, making for easier debugging of code and higher trust in the models. What does this mean for you? If you're a developer in the AI space, this will be big news for you. OpenAI took a long break from offering out its weights available to the public, and there is a clear shift in their thinking for this to become available. For everybody else, this won't be of much importance. The big update for the average person will be GPT-5 when that launches in the next week or so. OpenAI did promise a lot of big updates in the next few weeks, with this just being the starter for the main course soon to come. More from Tom's Guide ChatGPT-5 is coming — here's how it could change the way we prompt forever Amazon may bring ads to Alexa+ in least surprising move ever OpenAI says they are no longer optimizing ChatGPT to keep you chatting — here's why
Geeky Gadgets
06-08-2025
- Business
- Geeky Gadgets
OpenAI's Open Weight GPT-OSS 120B and 20B AI Models Fully Tested
What if the future of AI wasn't locked behind proprietary walls but instead placed in the hands of everyone? OpenAI's bold release of GPT-OSS 120B and GPT-OSS 20B, two new open weight models, has sparked a wave of excitement—and debate—across the AI community. With their unparalleled reasoning capabilities and scalability, these models promise to redefine what open source AI can achieve. Yet, their creative limitations and the shadow of leaked competitors like Horizon Alpha raise tough questions about innovation and accessibility. Are these models a true gift to the open source world, or do they fall short of their potential? In this overview, World of AI explore the unique strengths and trade-offs of OpenAI's GPT-OSS models, from their Apache 2.0 license freedoms to their performance in logic-driven tasks. You'll discover how these models stack up against alternatives, why their scalability makes them accessible to a diverse range of users, and where they might leave you wanting more. Whether you're a developer seeking cost-effective solutions or a researcher pushing the boundaries of AI, this deep dive will help you decide if OpenAI's latest release is the right fit for your needs. Sometimes, the most exciting innovations come with a twist—what will you make of this one? OpenAI's GPT-OSS AI Models Key Features and Specifications The GPT-OSS models are tailored to meet a variety of computational requirements, offering distinct advantages based on their scale and design. Below is a detailed comparison of their core features: GPT-OSS 120B: A large-scale model with 120 billion parameters, optimized for high-performance systems such as data centers and enterprise-level applications. A large-scale model with 120 billion parameters, optimized for high-performance systems such as data centers and enterprise-level applications. GPT-OSS 20B: A medium-scale model with 20 billion parameters, designed for broader accessibility, including use on desktops and laptops. A medium-scale model with 20 billion parameters, designed for broader accessibility, including use on desktops and laptops. Reasoning Capabilities: Both models excel in logical reasoning and mathematical problem-solving, using advanced chain-of-thought techniques. Both models excel in logical reasoning and mathematical problem-solving, using advanced chain-of-thought techniques. Context Length: Support for a 128k context length enables efficient handling of extensive text inputs, making them suitable for complex tasks. Support for a 128k context length enables efficient handling of extensive text inputs, making them suitable for complex tasks. Open source License: Distributed under Apache 2.0, allowing unrestricted experimentation, modification, and commercial use. These features make the GPT-OSS models versatile tools for developers and researchers, catering to a wide range of computational needs. Performance, Scalability, and Deployment The GPT-OSS models are designed to accommodate varying hardware capabilities, making sure scalability for users with different resource constraints. GPT-OSS 120B: This model is optimized for high-end systems, offering unparalleled performance for data-intensive tasks. With its 120 billion parameters, it is best suited for environments requiring significant computational power, such as enterprise-level data centers and advanced research facilities. This model is optimized for high-end systems, offering unparalleled performance for data-intensive tasks. With its 120 billion parameters, it is best suited for environments requiring significant computational power, such as enterprise-level data centers and advanced research facilities. GPT-OSS 20B: A more accessible alternative, this model is designed for everyday devices, including desktops and laptops. It provides a cost-effective solution for developers and researchers with limited hardware resources. Both models demonstrate exceptional reasoning capabilities, particularly in logic-driven tasks such as financial analysis, academic research, and mathematical computations. However, their creative outputs, including design and code generation, are inconsistent and often underperform when compared to other open source models. To enhance accessibility, OpenAI offers multiple deployment options: Local deployment for offline environments, making sure data privacy and security. API access for seamless integration into existing workflows and applications. Availability through platforms like Open Router, providing added flexibility for users. These deployment options, combined with the Apache 2.0 license, make the models adaptable to a wide range of use cases. OpenAI GPT-OSS AI Models 120B & 20B Tested Watch this video on YouTube. Take a look at other insightful guides from our broad collection that might capture your interest in OpenAI AI models. Technical Architecture and Cost Efficiency The GPT-OSS models employ OpenAI's proprietary training techniques and a mixture-of-experts architecture, which dynamically activates subsets of parameters to optimize efficiency. This architecture enables the models to balance performance with resource utilization, making them suitable for various applications. GPT-OSS 120B: Features 117 billion total parameters, with 5.1 billion active parameters per token, making sure high performance for complex tasks. Features 117 billion total parameters, with 5.1 billion active parameters per token, making sure high performance for complex tasks. GPT-OSS 20B: Comprises 21 billion total parameters, with 3.6 billion active parameters per token, offering a more resource-efficient alternative. The pricing structure for these models is designed to accommodate a variety of budgets, making them accessible to a broad audience: GPT-OSS 120B: Input tokens are priced at $0.15 per million, while output tokens cost $0.60 per million. Input tokens are priced at $0.15 per million, while output tokens cost $0.60 per million. GPT-OSS 20B: Input tokens are priced at $0.05 per million, with output tokens costing $0.20 per million. This token-based pricing model allows users to scale their usage based on specific requirements, making sure cost efficiency for both small-scale and large-scale applications. Strengths, Limitations, and Ideal Use Cases The GPT-OSS models are particularly well-suited for logic-intensive applications, offering robust performance in areas such as: Financial planning and analysis, where precise calculations and logical reasoning are critical. Academic research and data interpretation, allowing researchers to process and analyze large datasets efficiently. Offline AI applications, providing functionality in environments where internet access is unavailable. Despite their strengths, the models have notable limitations: Creative Performance: Their outputs in creative tasks, such as design and code generation, are inconsistent and often fall short of expectations when compared to other open source alternatives. Their outputs in creative tasks, such as design and code generation, are inconsistent and often fall short of expectations when compared to other open source alternatives. Content Restrictions: Fine-tuned to block malicious or restricted content, the models may lack adaptability in certain scenarios, limiting their flexibility for unrestricted applications. These limitations may affect their suitability for projects requiring high-quality creative outputs or unrestricted adaptability. However, their strengths in reasoning and logic-driven tasks make them valuable tools for developers and researchers. Impact on the Open source AI Community The release of the GPT-OSS 120B and 20B models represents a significant milestone for the open source AI community. By offering these models under the Apache 2.0 license, OpenAI has empowered developers and researchers to innovate without the constraints of proprietary systems. This move has the potential to foster greater collaboration and innovation within the AI ecosystem. However, the reception has been mixed. While many applaud OpenAI's commitment to open source accessibility, others express disappointment over unmet expectations, particularly when comparing these models to leaked alternatives like Horizon Alpha. This highlights the ongoing debate within the AI community regarding the balance between accessibility, performance, and innovation. The GPT-OSS models mark a pivotal step in the evolution of open weight AI, offering developers and researchers a robust foundation for a wide range of applications. Their strengths in reasoning and mathematical tasks, combined with their open source nature, make them valuable assets for advancing AI research and development. Media Credit: WorldofAI Filed Under: AI, Top News Latest Geeky Gadgets Deals Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.
