Meta Unveils Llama 3.3: A Powerful and Cost-Efficient AI Model

The new Llama 3.3 70B model has just dropped -- here's why it's a big deal

The new model is available for download and installation at Ollama, Hugging Face or at Meta's official Llama site. For developers, and those who want to use AI models on their own computers instead of the cloud, this is a big deal. Every new Llama release shows how small open models can compete and even beat the best of the rest. Meta has made no secret of the fact that it sees the open model paradigm as the best defense against potential abuse from proprietary products. Smaller models mean that users can use cheaper smaller graphics cards with less VRAM and still receive a decent performance from the AI. The key to the usability of AI on desktop computers lies in getting snappy responses. The best AI in the world is useless if it takes an hour to deliver an answer. By also giving users the ability to customize and enhance the base Llama models, there's also every chance that open will continue to keep pace with closed in the long run. The plan seems to be working since Llama models were downloaded over 20 million times this August alone, which is a 10x increase over the same time last year. A big factor behind these numbers is that each release of these Llama versions comes with a significant decrease in cost and increase in performance and capability. The new model supports eight languages, including Spanish, Hindi and Thai, and has been deliberately designed so developers can fine-tune and add on additional capabilities or languages as they need. Two points stand out from the success of these open models. First, there is a demographic of large and small companies that prefer to retain a measure of control over their AI product integration. There is also a growing group of AI enthusiasts and specialists who are looking to run smaller models on more modest consumer-level hardware. There are more than 60,000 derivative models on Hugging Face, showing the strength of demand for fine-tuning the Llama model. In addition, large enterprise users like Goldman Sachs, Accenture and Shopify are also using Llama internally. A lot of the large enterprise use is based on the cloud versions, whereas Llama is also building a sizable fan base for its more powerful models. Companies like Zoom and DoorDash, for instance, are using Llama as part of their AI mix in a wide variety of tasks, including customer support, software engineering and data analysis. This growing Llama ecosystem is a clever play by Meta. Not only does it establish the company's strength in general-purpose AI, but it also provides some strong marketing juice for its in-house Meta AI product. With over 350 million downloads of Llama models across the world to date, Meta has now grabbed a firm spot as one of the world's top AI companies. It's AI assistant, Meta AI, has just topped 600 million monthly users. This number is likely to explode once Llama 4 is released early next year as expected.

Meta unveils a new, more efficient Llama model | TechCrunch

Meta has announced the newest addition to its Llama family of generative AI models: Llama 3.3 70B. In a post on X, Ahmad Al-Dahle, VP of generative AI at Meta, said that the text-only Llama 3.3 70B delivers the performance of Meta's largest Llama model, Llama 3.1 405B, but more cheaply. "By leveraging the latest advancements in post-training techniques ... this model improves core performance at a significantly lower cost," Al-Dahle said. Ahmad Al-Dahle published a chart showing Llama 3.3 70B outperforming Google's Gemini 1.5 Pro, OpenAI's GPT-4o, and Amazon's newly released Nova Pro on a number of industry benchmarks, including MMLU, which evaluates a model's ability to understand and generate text. Via email, a Meta spokesperson said that the model should deliver improvements in areas like math, general knowledge, instruction following, and app use. Llama 3.3 70B, which is available for download from the AI dev platform Hugging Face and other sources, including the official Llama website, is Meta's latest play to dominate the AI field with "open" models that can be used and commercialized for a range of purposes. Meta's terms constrains how certain developers can use its Llama models; platforms with more than 700 million monthly users must request special permission from the company. But for many devs and companies, it's immaterial that Llama models aren't "open" in the strictest sense. According to Meta, its Llama models have racked up more than 650 million downloads. Meta has leveraged Llama for its own ends, as well. Meta AI, the company's AI assistant, which is powered entirely by Llama models, now has nearly 600 million monthly active users, according to an Instagram post by CEO Mark Zuckerberg on Friday. Zuckerberg claims that Meta AI is on track to be the most-used AI assistant in the world. The open nature of Llama has been a blessing and curse for Meta. In November, a report emerged that Chinese military researchers had used a Llama model to develop defense chatbot. Meta responded by making its Llama models available to U.S. defense partners. Meta has also expressed concerns about its ability to comply with the AI Act, the EU law that establishes a legal and regulatory framework for AI -- calling the law's implementation "too unpredictable." At issue for the company are related provisions in the GDPR, the EU's privacy law, pertaining to AI training. Meta trains AI models on the public data of Instagram and Facebook users who haven't opted out -- data that in Europe is subject to GDPR guarantees. EU regulators earlier this year requested that Meta halt training on European user data while they assessed the company's GDPR compliance. Meta relented, while at the same time endorsing an open letter calling for "a modern interpretation" of GDPR that doesn't "reject progress." Meta, not immune to the technical challenges other AI labs are facing, is ramping up its compute infrastructure to train and serve future generations of Llama models. The company announced Wednesday that it would build a $10 billion AI data center in Louisiana -- the largest AI data center it's ever built. Zuckerberg said on Meta's Q4 earnings call in August that to train the next major set of Llama models, Llama 4, the company will need 10x more compute than what was needed to train Llama 3. Training large language models can be a costly business. Meta's capital expenditures rose nearly 33% to $8.5 billion in Q2 2024, from $6.4 billion a year earlier, driven by investments in servers, data centers and network infrastructure.

Meta launches open source Llama 3.3, shrinking powerful bigger model into smaller size

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Meta's VP of generative AI, Ahmad Al-Dahle took to rival social network X today to announce the release of Llama 3.3, the latest open-source multilingual large language model (LLM) from the parent company of Facebook, Instagram, WhatsApp and Quest VR. As he wrote: "Llama 3.3 improves core performance at a significantly lower cost, making it even more accessible to the entire open-source community." With 70 billion parameters -- or settings governing the model's behavior -- Llama 3.3 delivers results on par with Meta's 405B parameter model from the Llama 3.1 from the summer, but at a fraction of the cost and computational overhead -- e.g., the GPU capacity needed to run the model in an inference. It's designed to offer top-tier performance and accessibility yet in a smaller package than prior foundation models. Meta's Llama 3.3 is offered under the Llama 3.3 Community License Agreement, which grants a non-exclusive, royalty-free license for use, reproduction, distribution, and modification of the model and its outputs. Developers integrating Llama 3.3 into products or services must include appropriate attribution, such as "Built with Llama," and adhere to an Acceptable Use Policy that prohibits activities like generating harmful content, violating laws, or enabling cyberattacks. While the license is generally free, organizations with over 700 million monthly active users must obtain a commercial license directly from Meta. A statement from the AI at Meta team underscores this vision: "Llama 3.3 delivers leading performance and quality across text-based use cases at a fraction of the inference cost." How much savings are we talkin' about, really? Some back-of-the-envelope math: Llama 3.1-405B requires between 243 GB and 1944 GB of GPU memory, according to the Substratus blog (for the open source cross cloud substrate). Meanwhile, the older Llama 2-70B requires between 42-168 GB of GPU memory, according to the same blog, though same have claimed as low as 4 GB, or as Exo Labs has shown, a few Mac computers with M4 chips and no discrete GPUs. Therefore, if the GPU savings for lower-parameter models holds up in this case, those looking to deploy Meta's most powerful open source Llama models can expect to save up to nearly 1940 GB worth of GPU memory, or potentially, 24 times reduced GPU load for a standard 80 GB Nvidia H100 GPU. At an estimated $25,000 per H100 GPU, that's up to $600,000 in up-front GPU cost savings, potentially -- not to mention the continuous power costs. A highly performant model in a small form factor According to Meta AI on X, the Llama 3.3 model handedly outperforms the identically sized Llama 3.1-70B as well as Amazon's new Nova Pro model in several benchmarks such as multilingual dialogue, reasoning, and other advanced natural language processing (NLP) tasks (Nova outperforms it in HumanEval coding tasks). Llama 3.3 has been pretrained on 15 trillion tokens from "publicly available" data and fine-tuned on over 25 million synthetically generated examples, according to the information Meta provided in the "model card" posted on its website. Leveraging 39.3 million GPU hours on H100-80GB hardware, the model's development underscores Meta's commitment to energy efficiency and sustainability. Llama 3.3 leads in multilingual reasoning tasks with a 91.1% accuracy rate on MGSM, demonstrating its effectiveness in supporting languages such as German, French, Italian, Hindi, Portuguese, Spanish, and Thai, in addition to English. Cost-effective and environmentally conscious Llama 3.3 is specifically optimized for cost-effective inference, with token generation costs as low as $0.01 per million tokens. This makes the model highly competitive against industry counterparts like GPT-4 and Claude 3.5, with greater affordability for developers seeking to deploy sophisticated AI solutions. Meta has also emphasized the environmental responsibility of this release. Despite its intensive training process, the company leveraged renewable energy to offset greenhouse gas emissions, resulting in net-zero emissions for the training phase. Location-based emissions totaled 11,390 tons of CO2-equivalent, but Meta's renewable energy initiatives ensured sustainability. Advanced features and deployment options The model introduces several enhancements, including a longer context window of 128k tokens (comparable to GPT-4o, about 400 pages of book text), making it suitable for long-form content generation and other advanced use cases. Its architecture incorporates Grouped Query Attention (GQA), improving scalability and performance during inference. Designed to align with user preferences for safety and helpfulness, Llama 3.3 uses reinforcement learning with human feedback (RLHF) and supervised fine-tuning (SFT). This alignment ensures robust refusals to inappropriate prompts and an assistant-like behavior optimized for real-world applications. Llama 3.3 is already available for download through Meta, Hugging Face, GitHub, and other platforms, with integration options for researchers and developers. Meta is also offering resources like Llama Guard 3 and Prompt Guard to help users deploy the model safely and responsibly.

Meta AI approaches 600 million monthly users with Llama 3.3

Meta has announced that its AI assistant, Meta AI, is approaching a significant milestone with nearly 600 million monthly users. CEO Mark Zuckerberg disclosed this update in a recent announcement. Originally launched last fall, Meta AI surpassed the 500 million user mark in October. The surge in user numbers coincides with the release of Meta's latest text model, Llama 3.3, which is designed to deliver high performance at a lower operational cost. Llama 3.3 introduces a new 70 billion parameter model that Meta claims mirrors the performance of its previous 405 billion parameter model while being more cost-effective to run. Ahmad Al-Dahle, Vice President of Generative AI at Meta, shared a performance comparison indicating that Llama 3.3 outperformed Google's Gemini Pro 1.5 and OpenAI's GPT-4o on several benchmarks. The model employs advancements in post-training techniques, including online preference optimization, marking a significant step forward in efficiency. Meta has officially released Llama 3.3 Zuckerberg provided a glimpse into the future of Meta's AI, revealing that Llama 4 is on the horizon. He indicated that the 3.3 version would be the "last big AI update of the year," hinting at forthcoming developments. While details about Llama 4 remain limited, Zuckerberg mentioned earlier this year that it was being trained using over 100,000 H100 GPUs, with an expected release for some "smaller" models planned for early next year.

Meta Launches New Llama 3.3 70B AI Model : Redefining Cost Efficiency

Meta has unveiled Llama 3.3, a 70-billion-parameter AI model that combines advanced capabilities with a focus on cost efficiency. This model is specifically designed to handle complex tasks such as long-context understanding, instruction following, and mathematical problem-solving. By offering a balance between high performance and affordability, Llama 3.3 provides developers with a powerful tool that minimizes operational expenses while requiring specialized hardware for optimal deployment. Llama 3.3 distinguishes itself by achieving a unique balance between size and performance, rivaling models with significantly larger parameter counts. Trained on a vast dataset of 15 trillion tokens with a knowledge cutoff of December 2023, it supports an extended context length of up to 128,000 tokens. This extended context capability enables the model to process and analyze large datasets or lengthy documents in a single pass, making it particularly well-suited for applications that demand detailed and nuanced long-context understanding. The model's design emphasizes cost-effectiveness, allowing for local deployment on developer workstations equipped with specialized hardware. This accessibility ensures that developers and businesses can use its capabilities without incurring the high costs typically associated with large-scale AI models. By combining efficiency with accessibility, Llama 3.3 positions itself as a practical and versatile solution for a wide range of AI-driven applications. Llama 3.3 delivers competitive performance across multiple domains, showcasing its versatility and advanced capabilities. Key highlights include: These performance metrics place Llama 3.3 among the top-performing AI models, competing directly with other advanced systems such as Gemini and Google's latest offerings. Its ability to deliver high-quality results across a variety of tasks underscores its value as a reliable and efficient AI solution. Here are more guides from our previous articles and guides related to Llama 3 that you may find helpful. One of the most compelling aspects of Llama 3.3 is its affordability, which sets it apart from many competitors. The model significantly reduces both input and output costs, making it an attractive option for businesses and developers seeking high-performance AI solutions without prohibitive expenses. Key cost metrics include: This dramatic reduction in operational costs makes Llama 3.3 a practical choice for organizations of all sizes, allowing them to integrate advanced AI capabilities into their workflows without exceeding budgetary constraints. By prioritizing cost efficiency, Meta has made innovative AI technology more accessible to a broader audience. While Llama 3.3 offers numerous advantages, it does come with specific technical requirements. The model is optimized for text-only applications, focusing on areas such as natural language processing, document analysis, and conversational systems. To achieve optimal performance, developers must use specialized hardware capable of handling the model's computational demands. Despite these requirements, accessibility is enhanced through its availability on popular hosting platforms such as Hugging Face and AMA. These platforms allow developers to easily download and experiment with the model, fostering innovation and allowing a wide range of use cases. This combination of technical sophistication and accessibility ensures that Llama 3.3 remains a practical choice for both research and commercial applications. Llama 3.3 has undergone rigorous independent benchmarking, demonstrating strong performance in key areas such as instruction following, code generation, and text-based tasks. These benchmarks validate its reliability and utility across a variety of applications. Additionally, several prominent hosting providers, including Deep Infra, Hyperbolic, Gro, Fireworks, and Together AI, have adopted the model, further highlighting its effectiveness and industry relevance. The model's ability to meet the demands of modern AI applications while maintaining cost efficiency makes it a valuable asset for businesses, researchers, and developers. Its adoption by leading hosting providers underscores its potential to drive innovation and streamline workflows across diverse sectors. The success of Llama 3.3 is rooted in Meta's advancements in alignment processes and reinforcement learning techniques. These innovations enhance the model's ability to follow instructions accurately and perform complex tasks with precision. By focusing on alignment, Meta has ensured that Llama 3.3 delivers reliable and consistent results across a wide range of applications, from academic research to commercial deployments. The integration of advanced alignment techniques also improves the model's capacity for nuanced understanding and context-aware responses. This focus on precision and reliability makes Llama 3.3 a versatile tool capable of addressing the challenges of modern AI applications while maintaining a high standard of performance. Llama 3.3 represents a significant advancement in AI development, combining a 70-billion-parameter architecture with extended context understanding and competitive performance. By bridging the gap between high capability and affordability, it sets a new benchmark for efficiency and accessibility in the AI landscape. Its ability to deliver advanced functionality at a fraction of the cost of competing models positions it as a fantastic tool for developers and businesses alike. With its focus on cost efficiency, technical sophistication, and practical applications, Llama 3.3 paves the way for more innovative and accessible AI solutions. As the demand for advanced AI technology continues to grow, Llama 3.3 stands out as a reliable and cost-effective option, driving progress and allowing new possibilities across industries.

Llama-3.3 70B Fully Tested : Open Source AI You Can't Ignore

This month Meta has introduced the Llama-3.3 70B, an advanced open-source large language model (LLM) that builds upon the foundation of its predecessor, Llama-3 70B. This release marks a significant step in the evolution of open-source artificial intelligence, offering improved performance, expanded functionality, and enhanced usability. Designed to address a diverse range of tasks, Llama-3.3 70B positions itself as a competitive alternative to proprietary models such as OpenAI's GPT-4 and Anthropic's Claude while maintaining the accessibility and adaptability that open-source software is known for. Llama-3.3 70B builds on the strengths of its predecessor, Llama-3 70B, with significant upgrades that promise to redefine what open-source AI can achieve. From excelling in benchmarks to supporting advanced features like function calling, this model is packed with potential. But what really sets it apart is its commitment to empowering users -- whether through its open-source licensing, its ability to handle diverse tasks, or its seamless integration into various platforms. Ready to dive into what makes this model a fantastic option? Let's explore how Llama-3.3 70B is reshaping the AI landscape. This isn't just another language model -- it's a leap forward in open-source AI, designed to make innovative technology more accessible, adaptable, and effective for everyone. If you've ever felt limited by proprietary systems or frustrated by models that just miss the mark, this might be the breakthrough you've been waiting for. Llama-3.3 70B distinguishes itself through its optimized Transformer architecture, which powers its autoregressive language modeling capabilities. The model has undergone fine-tuning using a combination of supervised learning and reinforcement learning with human feedback (RLHF). This dual approach enables it to generate more accurate and contextually relevant responses across a wide array of domains. Some of its standout features include: These features collectively enhance the model's utility, making it a valuable resource for developers, researchers, and organizations seeking a customizable and high-performing AI solution. Llama-3.3 70B has been rigorously tested using industry-standard benchmarks, including HumanEval, IIEval, and GPQ. These evaluations highlight its significant advancements over earlier Llama models and its ability to compete with leading proprietary alternatives. The model demonstrates exceptional performance in several areas: These results underscore the model's versatility and effectiveness, positioning it as a practical choice for applications ranging from software development to advanced data analysis. Gain further expertise in Llama 3 by checking out these recommendations. One of the most appealing aspects of Llama-3.3 70B is its accessibility. The model is available on multiple platforms, including Hugging Face, Together AI, Hyperbolic, and GHF, providing developers with the flexibility to deploy it in various environments. Whether you prefer cloud-based solutions or local deployment, the model's adaptability ensures smooth integration into diverse workflows. The open-source licensing further enhances its appeal by allowing users to modify and tailor the model to meet their specific needs. This approach aligns with Meta's commitment to fostering an open and collaborative AI ecosystem, empowering both individual developers and large organizations to innovate without the constraints of proprietary restrictions. When compared to proprietary models like GPT-4, Claude, and Quin 2.5, Llama-3.3 70B stands out as a robust and versatile alternative. While proprietary models may excel in certain specialized areas, Llama-3.3 70B's open-source nature and strong performance metrics make it an attractive option for those seeking a cost-effective and customizable solution. The model's ability to handle a wide range of tasks -- including programming, logical reasoning, and data analysis -- makes it a practical choice for both individual developers and enterprise-level applications. Its advancements over earlier iterations of the Llama series further highlight Meta's dedication to continuous improvement and innovation in the AI space. The Llama series is poised for further growth and innovation. Potential developments, such as the introduction of smaller models like 8B and 45B, could address specific performance gaps and expand the model's capabilities. These smaller iterations may also provide more resource-efficient options for users with limited computational power, broadening the accessibility of advanced AI tools. Collaborations with platforms like Hugging Face and Together AI are likely to enhance the model's integration and usability, fostering a more inclusive and dynamic open-source AI ecosystem. Such partnerships could pave the way for new applications and use cases, further solidifying Meta's leadership in the open-source LLM domain. Llama-3.3 70B represents a significant milestone in the development of open-source language models. With its optimized architecture, advanced fine-tuning techniques, and impressive performance across key benchmarks, it offers a compelling solution for a wide range of applications. By prioritizing accessibility, adaptability, and innovation, Meta continues to lead the charge in the open LLM space, empowering users to tackle complex challenges with confidence and efficiency.

Meta releases efficiency-optimized Llama 3.3 70B large language model - SiliconANGLE

Meta releases efficiency-optimized Llama 3.3 70B large language model Meta Platforms Inc. today introduced Llama 3.3 70B, the latest addition to its eponymous line of open-source large language models. The new algorithm provides similar output quality as Llama 3.1 405B, the most advanced LLM in the series, but using a fraction of the hardware. The result is a significant drop in infrastructure expenses. Meta says that Llama 3.3 70B generates prompt responses nearly five times more cost-efficiently. The model is based on an optimized version of the Transformer architecture, the neural network design that underpins most cutting-edge LLMs. When analyzing a set of data points, Transformer-based models use a so-called attention mechanism to determine which data points are most relevant to the task at hand. Meta swapped the default attention mechanism with an improved implementation that lowers inference costs. The company's engineers trained Llama 3.3 70B on a cluster of H100-80GB chips from Nvidia Corp. The chips' TDP, a metric that tracks the extent to which a processor's compute capacity is utilized, was set to the 700-watt maximum. Meta says that the LLM took 39.3 million graphics card hours to train. The training dataset includes about 15 trillion tokens, units of data that each correspond to a few letters or numbers. Meta used information from the public web, as well as more than 25 million synthetic examples. Those are AI-generated data points created specifically for LLM development purposes. After Meta completed the initial training process, it refined Llama 3.3 70B with several methods. One of the techniques the company used is known as supervised fine-tuning. It involves providing a freshly developed LLM with additional datasets that it didn't access during the initial training. Those additional datasets contain metadata, or contextual information, that makes it easier for the LLM to find useful patterns. Meta also used another AI method known as RLHF. While an LLM is being trained, it receives pointers from an algorithm on how to improve the quality of its output. RLHF combines those automatically-generated pointers with feedback from humans. After completing the development process, Meta compared Llama 3.3 70B with Llama 3.1 405B using ten AI benchmarks. Llama 3.3 70B trailed its larger namesake by under 2% in six of the tests and managed to achieve higher scores across three. It also mostly outperformed OpenAI's GPT-4o.

Llama 3.3 Just Made Synthetic Data Generation Effortless

Meta today unveiled Llama 3.3, a multilingual LLM to redefine AI's role in synthetic data generation. Featuring 70 billion parameters, Llama 3.3 is as performant as the previous 405B model yet optimised for efficiency and accessibility. Its multilingual output supports diverse languages, including Hindi, Portuguese, and Thai, empowering developers worldwide to create customised datasets for specialised AI models. "As we continue to explore new post-training techniques, today we're releasing Llama 3.3 -- a new open source model that delivers leading performance and quality across text-based use cases such as synthetic data generation at a fraction of the inference cost," shared Meta, on X. Developers can now use its expanded context length of 128k tokens to produce vast and high-quality datasets, addressing challenges like privacy restrictions and resource constraints. Meta's AI chief Yann LeCun previously said that this capability enables innovation in low-resource languages, a sentiment echoed by Indian entrepreneur Nandan Nilekani. "India should focus on building small, use-case-specific models quickly," Nilekani said, highlighting Llama's pivotal role in generating tailored training data for Indic language models. The success of such approaches is evident in projects like Sarvam AI's Sarvam 2B, which outperforms larger models in Indic tasks by utilising synthetic data generated with Llama. Hamid Shojanazeri, an ML engineer at Meta, said synthetic data generation solves critical bottlenecks in domains where collecting real-world datasets is too costly or infeasible. "Synthetic data is vital for advancing AI in privacy-sensitive areas or low-resource languages," he added. With its RLHF tuning and supervised fine-tuning, Llama 3.3 produces instruction-aligned datasets for tasks requiring high precision. Indic startups like Sarvam AI and Ola Krutrim have already reaped the benefits of Llama's capabilities. Sarvam AI's 2B model trained on 2 trillion synthetic Indic tokens demonstrates how such data can efficiently train smaller, purpose-built models while retaining high performance. "If you look at the 100 billion tokens in Indian languages, we used a clever method to create synthetic data for building these models using Llama 3.1 405B. We trained the model on 1,024 NVIDIA H100s in India, and it took only 15 days," said Sarvam AI chief Vivek Raghavan in an interview with AIM. Similarly, Llama 3.3's multilingual support and scalability make it indispensable for bridging the data divide in underrepresented languages. Llama 3.3's ability to support synthetic data generation extends beyond niche use cases, fostering broader adoption among developers, educators, and businesses. "By reducing the cost of producing high-quality training data, Llama accelerates innovation globally," said Ahmad Al-Dahle, Meta's VP of Generative AI. As speculation about GPT-4.5 intensifies, Llama 3.3 has decisively stepped in to meet immediate developer needs. With its revolutionary approach to synthetic data generation and cost-effectiveness, it's clear that Llama 3.3 isn't just filling a gap -- it's setting a new standard. "My synthetic data cost goes down 30x," said Pratik Desai, co-founder at KissanAI, on X. The release of Llama 3.3 fits squarely into Meta's long-term AI strategy. As Zuckerberg revealed during Meta's Q3 earnings call, the forthcoming Llama 4, set for early 2025, will introduce "new modalities, stronger reasoning, and much faster capabilities." This suggests that synthetic data generation capabilities refined in Llama 3.3 could become even more robust in future iterations. Meta's VP Ragavan Srinivasan recently hinted at advancements in "memory-based applications for coding and cross-modality support" for future Llama models. The robust framework established by Llama 3.3's synthetic data capabilities could be integral to these developments. By enabling developers to produce domain-specific training datasets, Meta positions itself as a critical enabler of innovation in both the private and public sectors. Future Llama versions will likely support an even broader array of languages and specialised use cases. As synthetic data generation becomes central to AI development, tools like Llama Guard 3 and enhanced tokenisation methods will ensure safe, responsible usage. For countries like India, where data creation in regional languages is critical, it offers an accessible pathway to developing culturally relevant AI solutions. Globally, as Mark Zuckerberg mentioned, Meta's next-generation data center in Louisiana promises to drive even more ambitious AI advancements: "We are in this for the long term, committed to building the most advanced AI in the world."

Meta Announces Llama 3.3 70B That Outperforms OpenAI's GPT-4o, Google's Gemini, and Amazon's AI Models - Alphabet (NASDAQ:GOOG), Amazon.com (NASDAQ:AMZN)

Mark Zuckerberg-led Meta Platforms, Inc. META has introduced Llama 3.3 70B, a new AI model that outperforms competitors like Alphabet Inc.'s GOOG GOOGL Google, OpenAI, and Amazon.com, Inc. AMZN. What Happened: Announced on Friday, the model offers the performance of Meta's largest Llama model, Llama 3.1 405B, but at a reduced cost. According to Ahmad Al-Dahle, Meta's VP of generative AI, Llama 3.3 70B leverages advanced post-training techniques to enhance core performance efficiently. Taking to X, formerly Twitter, he shared a chart demonstrating the model's superiority over Google's Gemini 1.5 Pro, OpenAI's GPT-4o, and Amazon's Nova Pro on various benchmarks, including MMLU. See Also: Amazon Halts Inferentia AI Chip Development To Take On Nvidia: How Trainium Is Shaping Up To Be The New Weapon In AI Chip Wars The model is available for download on platforms like Hugging Face and the official Llama website. In a post on Threads, Zuckerberg noted that Meta AI, powered by Llama models, now has nearly 600 million monthly active users. The company is also building a $10 billion AI data center in Louisiana to support future Llama models. Subscribe to the Benzinga Tech Trends newsletter to get all the latest tech developments delivered to your inbox. Why It Matters: Earlier in October it was reported that Meta plans to develop an AI-powered search engine to reduce its dependence on major tech players like Alphabet and Microsoft Corporation. Meta has also made a shift toward nuclear power to support its AI ambitions. The company is seeking proposals for nuclear energy projects to power its AI initiatives, aiming for 1-4 gigawatts of new nuclear generation capacity in the U.S. by the early 2030s. In its third-quarter earnings report, Meta reported a revenue beat of $40.59 billion, surpassing analyst expectations. The company also highlighted the strong momentum in AI. Price Action: Meta's stock rose 2.44% on Friday to close at $623.77 but dipped slightly by 0.08% in after-hours trading. So far this year, Meta shares have surged 80.13%, far outpacing the Nasdaq 100 index's 30.7% gain during the same timeframe, according to Benzinga Pro data. The consensus price target for Meta, based on evaluations from 41 analysts, stands at $639.05, with Rosenblatt setting the highest target at $811 on Oct. 31. Latest ratings from Raymond James, Wells Fargo, and Citigroup average $673.67, pointing to a potential upside of 8%. Check out more of Benzinga's Consumer Tech coverage by following this link. Read Next: Tesla CEO Elon Musk Agrees With Apple Co-Founder Steve Jobs On Guiding Talent: 'You Know Who The Best Managers Are?' Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors. Photo courtesy: Shutterstock Market News and Data brought to you by Benzinga APIs

Your PC Can't Handle Meta's New Llama AI Model (Probably)

Meta has released Llama 3.3 70B, a modified version of the company's most powerful AI model that can be downloaded to run on your own hardware. Your PC probably isn't ready for it, though. Like many other large language models (LLMs), Meta's Llama generative AI model is available in several parameter sizes for different use cases. For example, the smallest Llama 3.2 1B model can handle basic tasks with fast performance on the average smartphone, while the larger 11B and 90B versions are more powerful and need higher-end PCs and servers. The Llama models are primarily intended for text and chat functionality, but some versions can understand images too. Meta's new Llama 3.3 70B model is supposed to offer the same performance as the company's largest model, the 405B version, but with the ability to run on more PCs and servers. Meta's VP of generative AI said in a social media post, "By leveraging the latest advancements in post-training techniques including online preference optimization, this model improves core performance at a significantly lower cost." Even though this new 70B model is significantly shrunk down from the original 405B version, you'll still need a beefy PC or server to run it locally with acceptable performance. The file size is 37.14 GB, and LLMs generally need to fit in RAM to run well, so you'd probably need a machine with 64 GB RAM. You would also need a powerful GPU (or several paired together) for running the model. The model's description explains, "Llama 3.3 is intended for commercial and research use in multiple languages. Instruction tuned text only models are intended for assistant-like chat, whereas pretrained models can be adapted for a variety of natural language generation tasks. The Llama 3.3 model also supports the ability to leverage the outputs of its models to improve other models including synthetic data generation and distillation." Even though Llama 3.3 70B won't run on most computing hardware, you can run the smaller 1B, 3B, and 8B on many desktops and laptops with apps like LM Studio or Nvidia's Chat With RTX. My 16GB M1 Mac Mini runs Llama 3.1 8B at similar speeds as cloud-based AI chatbots, but I use smaller 3B models with my 8GB MacBook Air, since I have less RAM available. You can download Llama 3.3 70B and the other Lama models from Meta's website, Hugging Face, the built-in search in LM Studio, and other repositories. Source: TechCrunch

Meta Releases Smaller AI Model With Big Cost Savings | PYMNTS.com

Meta has unveiled a more efficient artificial intelligence (AI) model that could slash computing costs for businesses adopting AI technology. "Llama 3.3 improves core performance at a significantly lower cost, making it even more accessible to the entire open-source community," Meta's Vice President of Generative AI Ahmad Al-Dahle wrote in a Friday (Dec. 6) post on X. While Google and Amazon recently unveiled systems focused on enhanced capabilities like emotional understanding and video generation, Meta's new model reduces the computing power needed to run AI models to just 4 gigabytes of memory. The lower requirements could make advanced AI more accessible to smaller businesses, potentially saving money on hardware costs. Al-Dahle said the new model matches the capabilities of its other larger AI systems while using just 70 billion parameters, down from 405 billion in its predecessor. This reduction means companies could save up to $600,000 in hardware costs, as Llama 3.3 requires only 4 gigabytes of GPU memory compared to nearly 2,000 gigabytes for previous versions. Al-Dahle noted in his post that the model costs about $0.01 per million tokens to operate. It's being released as open-source software, though companies with more than 700 million monthly active users must obtain a commercial license. The model outperformed Amazon's Nova Pro -- released earlier this week as part of a new GenAI suite, with additional coverage below -- in multilingual dialogue and reasoning tasks, though Nova Pro maintains an advantage in coding tests. It achieved 91.1% accuracy on multilingual reasoning tasks, supporting languages including English, German, French, Italian, Hindi, Portuguese, Spanish and Thai. The release comes as tech companies race to reduce the computing resources needed for AI systems, a key factor in their commercial viability. Meta's shrinking model size while maintaining performance could make advanced AI models more accessible to smaller businesses. Experts previously told PYMNTS that open-source AI models are closing the gap between businesses and Big Tech's expensive systems, potentially bringing AI tools within the reach of smaller companies. Open source technology allows anyone to access, modify and share it. In other news, Google unveiled on Thursday (Dec. 5) PaliGemma 2, an AI system the company said can understand emotions and context in pictures. Unlike older systems that simply identify objects in photos, PaliGemma 2 can describe the emotional story behind an image. It comes in three sizes to fit different needs, and the largest version analyzes images using 28 billion parameters. "PaliGemma 2 generates detailed, contextually relevant captions for images, going beyond simple object identification to describe actions, emotions, and the overall narrative of the scene," Google wrote in a blog post. The system also shows promise in specialized tasks. It can interpret medical X-rays and recognize complex chemical formulas. Google has made PaliGemma 2 available to developers through popular AI platforms Hugging Face and Kaggle. Meanwhile, Amazon has unveiled a new suite of AI models called Nova, signaling its expanded presence in the AI market. Announced Tuesday (Dec. 3) at the AWS conference in Las Vegas, the platform includes Nova Reel for six-second video generation and Nova Canvas for text-to-image creation. The company said the new models offer improved speed, lower costs and fine-tuning capabilities. Nova Reel will soon support two-minute videos, while Canvas includes watermarking features to prevent misuse. "Inside Amazon, we have about 1,000 GenAI applications in motion, and we've had a bird's-eye view of what application builders are still grappling with," said Rohit Prasad, senior vice president of Amazon Artificial General Intelligence, in a news release. "Our new Amazon Nova models are intended to help with these challenges for internal and external builders, and provide compelling intelligence and content generation while also delivering meaningful progress on latency, cost-effectiveness, customization, information grounding, and agentic capabilities," Prasad added.